Yifu Chen, Shengpeng Ji, Zhengqing Liu, Qian Chen, Wen Wang, Ziqing Wang, Yangzhuo Li, Tianle Liang, Zhou Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

While diffusion and flow-matching models have advanced TTS, generating high-arousal emotions remains a persistent challenge due to the trade-off between stability and expressiveness.Existing systems often suffer from linguistic collapse when pursuing high intensity or fail to meet target emotional levels under stable settings.In this work, we identify that standard Gaussian initialization inevitably introduces a neutral prosody bias, while uniform Classifier-Free Guidance often distorts the acoustic manifold, leading to artifacts.To address this, we propose an inference framework that rectifies the emotional trajectory.An Emotion-Rectified Noise Prior injects a semantic gradient at initialization to align sampling with the target emotional manifold, and Likelihood-Inverse Guidance adaptively schedules guidance via a conditional/unconditional likelihood ratio, strengthening guidance only when the trajectory drifts toward a neutral fallback.Extensive experiments demonstrate that our method effectively resolves the stability bottleneck in high-intensity scenarios, achieving superior linguistic accuracy and emotional fidelity without model retraining.Code is available at https://github.com/MM-Speech/emo-tts.

Tianle Liang, Yifu Chen, Shengpeng Ji, Yijun Chen, Zhiyang Jia, Jingyu Lu, Fan Zhuo, Xueyi Pu, Yangzhuo Li, Zhou Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Jingyu Lu, Yuhan Wang, Fan Zhuo, Xize Cheng, Changhao Pan, Xueyi Pu, Yifu Chen, Chenyuhao Wen, Tianle Liang, Zhou Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Sashuai Zhou, Qiang Zhou, Jijin Hu, Hanqing Yang, Yue Cao, Junpeng Ma, Yinchao Ma, Jun Song, Tiezheng Ge, Cheng Yu, Bo Zheng, Zhou Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Knowledge-based Visual Question Answering (VQA) involves answering questions based not only on the given image, but also on external knowledge. Existing methods for knowledge-based VQA can be classified into two main categories: those that rely on external knowledge bases, and those that use Large Language Models (LLMs) as implicit knowledge engines. However, the former approach heavily relies on the quality of information retrieval, introducing additional information bias to the entire system. And the latter approach suffers from the extremely high computational cost and the loss of image information. To address these issues, we propose a novel framework called TAG that reformulates knowledge-based VQA as a contrastive learning problem. We innovatively propose a triple asymmetric paradigm, which aligns a lightweight text encoder to the image space with an extremely low training cost (0.0152B trainable parameters), and enhance its understanding ability on semantic granularity. TAG is both computation-efficient and effective, and we evaluate it on the knowledge-based VQA datasets, A-OKVQA, OK-VQA and VCR. The results show that TAG (0.387B) achieves the state-of-the-art performance when compared to methods using less than 1B parameters. Besides, TAG still shows competitive performance when compared to methods with LLM.

In music streaming applications, next music tracks are typically played automatically and require less consumption time than many other contents (e.g., books and movies). These characteristics potentially lead to long and noisy user listening history, making generic sequential recommendation models more error-prone when modeling user interests. To bridge the gap, we propose to retrieve representative music tracks from the original listening history for recommendation. Differing from existing user behavior retrieval techniques, we pursue a fine-grained, content-based, and temporal user behavior retrieval framework, namely FactUBR, to better exploit music track content and the fine-grained connections between temporally dependent tracks in the listening history. Technically, FactUBR consists of a reinforced content-based retrieval module (RCB) and a differentiable temporal-channel (DTC) retrieval module. RCB follows a reinforcement-learning procedure to optimize sequential retrieval decisions based on content differences between the state and the observation, maximizing rewards of retrieval diversity and recommendation performance. DTC evaluates the fine-grained correlations between temporal listening segments and the candidate music track, and employs the perturbed maximum technique for hard retrieval optimization. Extensive experiments on two public music recommendation benchmarks demonstrate that FactUBR can enhance various representative sequential recommendation models and outperform state-of-the-art (SOTA) behavior retrieval techniques.

Singing voice synthesis has made remarkable progress in generating natural and high-quality voices. However, existing methods rarely provide precise control over vocal techniques such as intensity, mixed voice, falsetto, bubble, and breathy tones, thus limiting the expressive potential of synthetic voices. We introduce TechSinger, an advanced system for controllable singing voice synthesis that supports five languages and seven vocal techniques. TechSinger leverages a flow-matching-based generative model to produce singing voices with enhanced expressive control over various techniques. To enhance the diversity of training data, we develop a technique detection model that automatically annotates datasets with phoneme-level technique labels. Additionally, our prompt-based technique prediction model enables users to specify desired vocal attributes through natural language, offering fine-grained control over the synthesized singing. Experimental results demonstrate that TechSinger significantly enhances the expressiveness and realism of synthetic singing voices, outperforming existing methods in terms of audio quality and technique-specific control.

Speech watermarking techniques can proactively mitigate the potential harmful consequences of instant voice cloning techniques. These techniques involve the insertion of signals into speech that are imperceptible to humans but can be detected by algorithms. Previous approaches typically embed watermark messages into continuous space. However, intuitively, embedding watermark information into robust discrete latent space can significantly improve the robustness of watermarking systems. In this paper, we propose DiscreteWM, a novel speech watermarking framework that injects watermarks into the discrete intermediate representations of speech. Specifically, we map speech into discrete latent space with a vector-quantized autoencoder and inject watermarks by changing the modular arithmetic relation of discrete IDs. To ensure the imperceptibility of watermarks, we also propose a manipulator model to select the candidate tokens for watermark embedding. Experimental results demonstrate that our framework achieves state-of-the-art performance in robustness and imperceptibility, simultaneously. Moreover, our flexible frame-wise approach can serve as an efficient solution for both voice cloning detection and information hiding. Additionally, DiscreteWM can encode 1 to 150 bits of watermark information within a 1-second speech clip, indicating its encoding capacity.

In this study, we focus on heterogeneous knowledge transfer across entirely different model architectures, tasks, and modalities. Existing knowledge transfer methods (e.g., backbone sharing, knowledge distillation) often hinge on shared elements within model structures or task-specific features/labels, limiting transfers to complex model types or tasks. To overcome these challenges, we present MergeNet, which learns to bridge the gap of parameter spaces of heterogeneous models, facilitating the direct interaction, extraction, and application of knowledge within these parameter spaces. The core mechanism of MergeNet lies in the parameter adapter, which operates by querying the source model's low-rank parameters and adeptly learning to identify and map parameters into the target model. MergeNet is learned alongside both models, allowing our framework to dynamically transfer and adapt knowledge relevant to the current stage, including the training trajectory knowledge of the source model. Extensive experiments on heterogeneous knowledge transfer demonstrate significant improvements in challenging settings, where representative approaches may falter or prove less applicable.

Zehan Wang, Ke Lei, Chen Zhu, Jiawei Huang, Sashuai Zhou, Luping Liu, Xize Cheng, Shengpeng Ji, Zhenhui Ye, Tao Jin, Zhou Zhao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Minghui Fang, Shengpeng Ji, Jialong Zuo, Hai Huang, Yan Xia, Jieming Zhu, Xize Cheng, Xiaoda Yang, Wenrui Liu, Gang Wang, Zhenhua Dong, Zhou Zhao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Xueyu Hu, Tao Xiong, Biao Yi, Zishu Wei, Ruixuan Xiao, Yurun Chen, Jiasheng Ye, Meiling Tao, Xiangxin Zhou, Ziyu Zhao, Yuhuai Li, Shengze Xu, Shenzhi Wang, Xinchen Xu, Shuofei Qiao, Zhaokai Wang, Kun Kuang, Tieyong Zeng, Liang Wang, Jiwei Li, Yuchen Eleanor Jiang, Wangchunshu Zhou, Guoyin Wang, Keting Yin, Zhou Zhao, Hongxia Yang, Fan Wu, Shengyu Zhang, Fei Wu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

In recent years, large language models have achieved significant success in generative tasks (e.g., speech cloning and audio generation) related to speech, audio, music, and other signal domains. A crucial element of these models is the discrete acoustic codecs, which serve as an intermediate representation replacing the mel-spectrogram. However, there exist several gaps between discrete codecs and downstream speech language models. Specifically, 1) Due to the reconstruction paradigm of the Codec model and the structure of residual vector quantization, the initial channel of the codebooks contains excessive information, making it challenging to directly generate acoustic tokens from weakly supervised signals such as text in downstream tasks. 2) Achieving good reconstruction performance requires the utilization of numerous codebooks, which increases the burden on downstream speech language models. Consequently, leveraging the characteristics of speech language models, we propose Language-Codec. In the Language-Codec, we introduce a Masked Channel Residual Vector Quantization (MCRVQ) mechanism along with improved fourier transform structures, refined discriminator design to address the aforementioned gaps. We compare our method with competing audio compression algorithms and observe significant outperformance across extensive evaluations. Furthermore, we also validate the efficiency of the Language-Codec on downstream speech language models. The source code and pretrained models will be open-sourced after the paper is accepted. Codes are available at https://github.com/jishengpeng/Languagecodec.

In this paper, we present ControlSpeech, a text-to-speech (TTS) system capable of fully cloning the speaker’s voice and enabling arbitrary control and adjustment of speaking style. Prior zero-shot TTS models only mimic the speaker’s voice without further control and adjustment capabilities while prior controllable TTS models cannot perform speaker-specific voice generation. Therefore, ControlSpeech focuses on a more challenging task—a TTS system with controllable timbre, content, and style at the same time. ControlSpeech takes speech prompts, content prompts, and style prompts as inputs and utilizes bidirectional attention and mask-based parallel decoding to capture codec representations corresponding to timbre, content, and style in a discrete decoupling codec space. Moreover, we analyze the many-to-many issue in textual style control and propose the Style Mixture Semantic Density (SMSD) module, which is based on Gaussian mixture density networks, to resolve this problem. To facilitate empirical validations, we make available a new style controllable dataset called VccmDataset. Our experimental results demonstrate that ControlSpeech exhibits comparable or state-of-the-art (SOTA) performance in terms of controllability, timbre similarity, audio quality, robustness, and generalizability. Codes are available at https://github.com/jishengpeng/ControlSpeech.

Recent advancements in latent diffusion models (LDMs) have markedly enhanced text-to-audio generation, yet their iterative sampling processes impose substantial computational demands, limiting practical deployment. While recent methods utilizing consistency-based distillation aim to achieve few-step or single-step inference, their one-step performance is constrained by curved trajectories, preventing them from surpassing traditional diffusion models. In this work, we introduce FlashAudio with rectified flows to learn straight flow for fast simulation. To alleviate the inefficient timesteps allocation and suboptimal distribution of noise, FlashAudio optimizes the time distribution of rectified flow with Bifocal Samplers and proposes immiscible flow to minimize the total distance of data-noise pairs in a batch vias assignment. Furthermore, to address the amplified accumulation error caused by the classifier-free guidance (CFG), we propose Anchored Optimization, which refines the guidance scale by anchoring it to a reference trajectory. Experimental results on text-to-audio generation demonstrate that FlashAudio’s one-step generation performance surpasses the diffusion-based models with hundreds of sampling steps on audio quality and enables a sampling speed of 400x faster than real-time on a single NVIDIA 4090Ti GPU. Code will be available at https://github.com/liuhuadai/FlashAudio. Audio Samples are available at https://FlashAudio-TTA.github.io/.

Jialong Zuo, Shengpeng Ji, Minghui Fang, Mingze Li, Ziyue Jiang, Xize Cheng, Xiaoda Yang, Chen Feiyang, Xinyu Duan, Zhou Zhao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

In recent years, speech generation fields have achieved significant advancements, primarily due to improvements in large TTS (text-to-speech) systems and scalable TTS datasets. However, there is still a lack of large-scale multilingual TTS datasets, which limits the development of cross-language and multilingual TTS systems. Hence, we refine Voxpopuli dataset and propose VoxpopuliTTS dataset. This dataset comprises 30,000 hours of high-quality speech data, across 3 languages with multiple speakers and styles, suitable for various speech tasks such as TTS and ASR. To enhance the quality of speech data from Voxpopuli, we improve the existing processing pipeline by: 1) filtering out low-quality speech-text pairs based on ASR confidence scores, and 2) concatenating short transcripts by checking semantic information completeness to generate the long transcript. Experimental results demonstrate the effectiveness of the VoxpopuliTTS dataset and the proposed processing pipeline.

Contrastive Language-Image Pre-training (CLIP) learns robust visual models through language supervision, making it a crucial visual encoding technique for various applications. However, CLIP struggles with comprehending spatial concepts in images, potentially restricting the spatial intelligence of CLIP-based AI systems. In this work, we propose SpatialCLIP, an enhanced version of CLIP with better spatial understanding capabilities. To capture the intricate 3D spatial relationships in images, we improve both "visual model" and "language supervision" of CLIP. Specifically, we design 3D-inspired ViT to replace the standard ViT in CLIP. By lifting 2D image tokens into 3D space and incorporating design insights from point cloud networks, our visual model gains greater potential for spatial perception. Meanwhile, captions with accurate and detailed spatial information are very rare. To explore better language supervision for spatial understanding, we re-caption images and perturb their spatial phrases as negative descriptions, which compels the visual model to seek spatial cues to distinguish these hard negative captions. With the enhanced visual model, we introduce SpatialLLaVA, following the same LLaVA-1.5 training protocol, to investigate the importance of visual representations for MLLM’s spatial intelligence. Furthermore, we create SpatialBench, a benchmark specifically designed to evaluate CLIP and MLLM in spatial reasoning. Spatial-CLIP and SpatialLLaVA achieve substantial performance improvements, demonstrating stronger capabilities in spatial perception and reasoning, while maintaining comparable results on general-purpose benchmarks.

Recent advancements in robotic manipulation have high- lighted the potential of intermediate representations for improving policy generalization. In this work, we explore grounding masks as an effective intermediate representation, balancing two key advantages: (1) effective spatial guidance that specifies target objects and placement areas while also conveying information about object shape and size, and (2) broad generalization potential driven by large-scale vision-language models pretrained on diverse grounding datasets. We introduce ROBOGROUND, a grounding-aware robotic manipulation policy that leverages grounding masks as an intermediate representation to guide policy networks in object manipulation tasks. To further explore and enhance generalization, we propose an automated pipeline for generating large-scale, simulated data with a diverse set of objects and instructions. Extensive experiments show the value of our dataset and the effectiveness of grounding masks as intermediate guidance, significantly enhancing the generalization abilities of robot policies. Code and data will be available at robo-ground.github.io.

Large language models (LLMs) have significantly enhanced cross-modal understanding capabilities by integrating visual encoders with textual embeddings, giving rise to multimodal large language models (MLLMs). However, these models struggle with non-natural images such as geometric and charts, particularly in fields like education and finance. Despite efforts to collect datasets and fine-tune the MLLMs, the gap with natural image understanding is still evident, and the cost of collecting large and diverse non-natural image datasets is high. To address this, we analyzed the limitations of transformer-based vision encoders(ViT) within existing MLLMs from a frequency perspective. Studies have shown that ViT models are less effective at capturing high-frequency information, impairing their ability to capture elements like points, lines, and angles in non-natural images. In response, we introduced FM-ViT, a frequency-modulated vision encoder that utilizes Fourier decomposition to extract high and low frequency components from self-attention features and re-weight them during tuning to non-natural images. In addition, we combine the features of CNN models with FM-ViT and propose EDGE, an MLLM with enhanced graphical encoders tailored for understanding non-natural images. Extensive experiments have confirmed the effectiveness of our FM-ViT and EDGE in 4 types. Project page.

We introduce a novel, training-free approach for enhancing alignment in Transformer-based Text-Guided Diffusion Models (TGDMs). Existing TGDMs often struggle to generate semantically aligned images, particularly when dealing with complex text prompts or multi-concept attribute binding challenges. Previous U-Net-based methods primarily optimized the latent space, but their direct application to Transformer-based architectures has shown limited effectiveness. Our method addresses these challenges by directly optimizing cross-attention maps during the generation process. Specifically, we introduce Self-Coherence Guidance, a method that dynamically refines attention maps using masks derived from previous denoising steps, ensuring precise alignment without additional training. To validate our approach, we constructed more challenging benchmarks for evaluating coarse-grained attribute binding, fine-grained attribute binding, and style binding. Experimental results demonstrate the superior performance of our method, significantly surpassing other state-of-the-art methods across all evaluated tasks. Our code is available at https://scg-diffusion.github.io/scg-diffusion.

Diffusion-based audio-driven talking avatar methods have recently gained attention for their high-fidelity, vivid, and expressive results. However, their slow inference speed limits practical applications. Despite the development of various distillation techniques for diffusion models, we found that naive diffusion distillation methods do not yield satisfactory results. Distilled models exhibit reduced robustness with open-set input images and a decreased correlation between audio and video compared to teacher models, undermining the advantages of diffusion models. To address this, we propose FADA (Fast Diffusion Avatar Synthesis with Mixed-Supervised Multi-CFG Distillation). We first designed a mixed-supervised loss to leverage data of varying quality and enhance the overall model capability as well as robustness. Additionally, we propose a multi-CFG distillation with learnable tokens to utilize the correlation between audio and reference image conditions, reducing the threefold inference runs caused by multi-CFG with acceptable quality degradation. Extensive experiments across multiple datasets show that FADA generates vivid videos comparable to recent diffusion model-based methods while achieving an NFE speedup of 4.17-12.5 times. Demos are available at our webpage https://fadavatar.github.io.

Zhipeng Bian, Jieming Zhu, Qijiong Liu, Wang Lin, Guohao Cai, Zhaocheng Du, Jiacheng Sun, Zhou Zhao, Zhenhua Dong. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Sashuai Zhou, Weinan Gan, Qijiong Liu, Ke Lei, Jieming Zhu, Hai Huang, Yan Xia, Ruiming Tang, Zhenhua Dong, Zhou Zhao. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recently, audio-visual speech recognition has attracted increasing attention. However, most existing works only focused on scenarios with two speakers. In this work, we study the effect of speaker number in AVSR task and propose an end-to-end audio-visual speech recognition framework under a more realistic condition where the speaker number is arbitrary. Specifically, we adopted curriculum learning to train models from easy scenarios to hard ones and introduce a new training strategy named Challenge-based Curriculum Learning (CBCL) that forces the model to focus on hard, challenging data instead of easy ones during training. Further, to avoid scenario bias from unbalanced sampling during curriculum learning, we propose a Speaker-number Aware Mixture-of-Expert (SA-MoE) mechanism to explicitly model the characteristic difference in scenarios with different speaker numbers.

3D spatial sound field synthesis takes the head-mounted audio signals and body poses as input and renders a 3D sound field around the center body, in which spatial audio can be inferred at any arbitrary position. To achieve this, a multi-modal system is required to spatialize input audio signals with the guidance of pose streams. Then, a numerical method is used to render the sound field with hundreds of spatial audio. However, there exist several challenges including hybrid audio input signals, cross-modal matching, and few data resources. To address them, we propose NAT3DSound, a unit-based framework for high-quality 3D spatial sound synthesis, which consists of 1) a general tokenization scheme for both dense and impulsive input audio signals, 2) a non-autoregressive transformer with multi-scale modality fusion for efficient cross-modal alignment and 3) a parallel sampling strategy for fast prediction. Furthermore, we investigate the feasibility of acoustic pre-training for low-resource learning in data-scarce scenarios. Extensive experiments and ablation studies demonstrate the effectiveness of NAT3DSound in terms of spatialization quality and generalization ability. Audio samples are available at http://NAT3DSound.github.io

This paper introduces PFlow-VC, a conditional flow matching voice conversion model that leverages fine-grained discrete pitch tokens and target speaker prompt information for expressive voice conversion (VC). Previous VC works primarily focus on speaker conversion, with further exploration needed in enhancing expressiveness (such as prosody and emotion) for timbre conversion. Unlike previous methods, we adopt a simple and efficient approach to enhance the style expressiveness of voice conversion models. Specifically, we pretrain a self-supervised pitch VQVAE model to discretize speaker-irrelevant pitch information and leverage a masked pitch-conditioned flow matching model for Mel-spectrogram synthesis, which provides in-context pitch modeling capabilities for the speaker conversion model, effectively improving the voice style transfer capacity. Additionally, we improve timbre similarity by combining global timbre embeddings with time-varying timbre tokens. Experiments on unseen LibriTTS test-clean and emotional speech dataset ESD show the superiority of the PFlow-VC model in both timbre conversion and style transfer. Audio samples are available on the demo page https://speechai-demo.github.io/PFlow-VC/.

This paper extends Cross Modal Generalization (CMG) to open-set environments by proposing the more challenging Open-set Cross Modal Generalization (OSCMG) task. This task evaluates multimodal unified representations in open-set conditions, addressing the limitations of prior closed-set cross-modal evaluations. OSCMG requires not only cross-modal knowledge transfer but also robust generalization to unseen classes within new modalities, a scenario frequently encountered in real-world applications. Existing multimodal unified representation work lacks consideration for open-set environments. To tackle this, we propose MICU, comprising two key components: Fine-Coarse Masked multimodal InfoNCE (FCMI) and Cross modal Unified Jigsaw Puzzles (CUJP). FCMI enhances multimodal alignment by applying contrastive learning at both holistic semantic and temporal levels, incorporating masking to enhance generalization. CUJP enhances feature diversity and model uncertainty by integrating modality-agnostic feature selection with self-supervised learning, thereby strengthening the model's ability to handle unknown categories in open-set tasks. Extensive experiments on CMG and the newly proposed OSCMG validate the effectiveness of our approach. The code is available at https://github.com/haihuangcode/CMG.

Domain Generalization (DG) aims to enhance model robustness in unseen or distributionally shifted target domains through training exclusively on source domains. Although existing DG techniques, such as data manipulation, learning strategies, and representation learning, have shown significant progress, they predominantly address single-modal data. With the emergence of numerous multi-modal datasets and increasing demand for multi-modal tasks, a key challenge in Multi-modal Domain Generalization (MMDG) has emerged: enabling models trained on multi-modal sources to generalize to unseen target distributions within the same modality set. Due to the inherent differences between modalities, directly transferring methods from single-modal DG to MMDG typically yields sub-optimal results. These methods often exhibit randomness during generalization due to the invisibility of target domains and fail to consider inter-modal consistency. Applying these methods independently to each modality in the MMDG setting before combining them can lead to divergent generalization directions across different modalities, resulting in degraded generalization capabilities. To address these challenges, we propose a novel approach that leverages Unified Representations to map different paired modalities together, effectively adapting DG methods to MMDG by enabling synchronized multi-modal improvements within the unified space. Additionally, we introduce a supervised disentanglement framework that separates modal-general and modal-specific information, further enhancing the alignment of unified representations. Extensive experiments on benchmark datasets, including EPIC-Kitchens and Human-Animal-Cartoon, demonstrate the effectiveness and superiority of our method in enhancing multi-modal domain generalization.

Recently, human-computer interaction with various modalities has shown promising applications, like GPT-4o and Gemini. Meanwhile, multimodal representation models have emerged as the foundation for these versatile multimodal understanding and generation pipeline. Models like CLIP, CLAP and ImageBind can map their specialized modalities into respective joint spaces. To construct a high-quality omni representation space that can be shared and expert in any modality, we propose to merge these advanced models into a unified space in scale. With this insight, we present \textbf{OmniBind}, advanced multimodal joint representation models via fusing knowledge of 14 pre-trained spaces, which support 3D, audio, image, video and language inputs. To alleviate the interference between different knowledge sources in integrated space, we dynamically assign weights to different spaces by learning routers with two objectives: cross-modal overall alignment and language representation decoupling. Notably, since binding and routing spaces only require lightweight networks, OmniBind is extremely training-efficient. Extensive experiments demonstrate the versatility and superiority of OmniBind as an omni representation model, highlighting its great potential for diverse applications, such as any-query and composable multimodal understanding.

With the rapid advancement of large models, voice assistants are gradually acquiring the ability to engage in open-ended daily conversations with humans. However, current spoken dialogue systems often overlook multi-modal information in audio beyond text, such as speech rate, volume, emphasis, and background sounds. Relying solely on Automatic Speech Recognition (ASR) can lead to the loss of valuable auditory cues, thereby weakening the system’s ability to generate contextually appropriate responses. To address this limitation, we propose \textbf{VoxDialogue}, a comprehensive benchmark for evaluating the ability of spoken dialogue systems to understand multi-modal information beyond text. Specifically, we have identified 12 attributes highly correlated with acoustic information beyond words and have meticulously designed corresponding spoken dialogue test sets for each attribute, encompassing a total of 4.5K multi-turn spoken dialogue samples. Finally, we evaluated several existing spoken dialogue models, analyzing their performance on the 12 attribute subsets of VoxDialogue. Experiments have shown that in spoken dialogue scenarios, many acoustic cues cannot be conveyed through textual information and must be directly interpreted from the audio input. In contrast, while direct spoken dialogue systems excel at processing acoustic signals, they still face limitations in handling complex dialogue tasks due to their restricted context understanding capabilities. All data and code will be open source at \url{https://voxdialogue.github.io/}.

Query-based sound separation (QSS) effectively isolate sound signals that match the content of a given query, enhancing the understanding of audio data. However, most existing QSS methods rely on a single modality for separation, lacking the ability to fully leverage homologous but heterogeneous information across multiple modalities for the same sound signal. To address this limitation, we introduce Omni-modal Sound Separation (**OmniSep**), a novel framework capable of isolating clean soundtracks based on omni-modal queries, encompassing both single-modal and multi-modal composed queries. Specifically, we introduce the **Query-Mixup** strategy, which blends query features from different modalities during training. This enables OmniSep to optimize multiple modalities concurrently, effectively bringing all modalities under a unified framework for sound separation. We further enhance this flexibility by allowing queries to influence sound separation positively or negatively, facilitating the retention or removal of specific sounds as desired. Finally, OmniSep employs a retrieval-augmented approach known as **Query-Aug**, which enables open-vocabulary sound separation. Experimental evaluations on MUSIC, VGGSOUND-CLEAN+, and MUSIC-CLEAN+ datasets demonstrate effectiveness of OmniSep, achieving state-of-the-art performance in text-, image-, and audio-queried sound separation tasks. For samples and further information, please visit the demo page at \url{https://omnisep.github.io/}.

Language models have been effectively applied to modeling natural signals, such as images, video, speech, and audio. A crucial component of these models is the codec tokenizer, which compresses high-dimensional natural signals into lower-dimensional discrete tokens. In this paper, we introduce WavTokenizer, which offers several advantages over previous SOTA acoustic codec models in the audio domain: 1) extreme compression. By compressing the layers of quantizers and the temporal dimension of the discrete codec, one-second audio of 24kHz sampling rate requires only a single quantizer with 40 or 75 tokens. 2) improved subjective quality. Despite the reduced number of tokens, WavTokenizer achieves state-of-the-art reconstruction quality with outstanding UTMOS scores and inherently contains richer semantic information. Specifically, we achieve these results by designing a broader VQ space, extended contextual windows, and improved attention networks, as well as introducing a powerful multi-scale discriminator and an inverse Fourier transform structure. We conducted extensive reconstruction experiments in the domains of speech, audio, and music. WavTokenizer exhibited strong performance across various objective and subjective metrics compared to state-of-the-art models. We also tested semantic information, VQ utilization, and adaptability to generative models. Comprehensive ablation studies confirm the necessity of each module in WavTokenizer. The code is available at https://github.com/jishengpeng/WavTokenizer.

Speech-driven 3D facial animation has attracted significant attention due to its wide range of applications in animation production and virtual reality. Recent research has explored speech-emotion disentanglement to enhance facial expressions rather than manually assigning emotions. However, this approach face issues such as feature confusion, emotions weakening and mean-face. To address these issues, we present EcoFace, a framework that (1) proposes a novel collaboration objective to provide a explicit signal for emotion representation learning from the speaker's expressive movements and produced sounds, constructing an audio-visual joint and coordinated emotion space that is independent of speech content. (2) constructs a universal facial motion distribution space determined by speech features and implement speaker-specific generation. Extensive experiments show that our method achieves more generalized and emotionally realistic talking face generation compared to previous methods.

Language Models (LMs) are increasingly used for type inference, aiding in error detection and software development. Some real-world deployments of LMs require the model to run on local machines to safeguard the intellectual property of the source code. This setting often limits the size of the LMs that can be used. We present Nester, the first neuro-symbolic approach that enhances LMs for type inference by integrating symbolic learning without increasing model size. Nester breaks type inference into sub-tasks based on the data and control flow of the input code, encoding them as a modular high-level program. This program executes multi-step actions, such as evaluating expressions and analyzing conditional branches of the target code, combining static typing with LMs to infer potential types. Evaluated on the ManyTypes4Py dataset in Python, Nester outperforms two state-of-the-art type inference methods (HiTyper and TypeGen), achieving 70.7\% Top-1 Exact Match, which is 18.3\% and 3.6\% higher than HiTyper and TypeGen, respectively. For complex type annotations like typing.Optional and typing.Union, Nester achieves 51.0\% and 16.7\%, surpassing TypeGen by 28.3\% and 5.8\%.

Orientation is a fundamental attribute of objects, essential for understanding their spatial pose and arrangement. However, practical solutions for estimating the orientation of open-world objects in monocular images remain underexplored. In this work, we introduce Orient Anything, the first foundation model for zero-shot object orientation estimation. A key challenge in this task is the scarcity of orientation annotations for open-world objects. To address this, we propose leveraging the vast resources of 3D models. By developing a pipeline to annotate the front face of 3D objects and render them from random viewpoints, we curate 2 million images with precise orientation annotations across a wide variety of object categories. To fully leverage the dataset, we design a robust training objective that models the 3D orientation as probability distributions over three angles and predicts the object orientation by fitting these distributions. Besides, we propose several strategies to further enhance the synthetic-to-real transfer. Our model achieves state-of-the-art orientation estimation accuracy on both rendered and real images, demonstrating impressive zero-shot capabilities across various scenarios. Furthermore, it shows great potential in enhancing high-level applications, such as understanding complex spatial concepts in images and adjusting 3D object pose.

Traditional video-to-audio generation techniques primarily focus on perspective video and non-spatial audio, often missing the spatial cues necessary for accurately representing sound sources in 3D environments. To address this limitation, we introduce a novel task, 360V2SA, to generate spatial audio from 360-degree videos, specifically producing First-order Ambisonics (FOA) audio - a standard format for representing 3D spatial audio that captures sound directionality and enables realistic 3D audio reproduction. We first create Sphere360, a novel dataset tailored for this task that is curated from real-world data. We also design an efficient semi-automated pipeline for collecting and cleaning paired video-audio data. To generate spatial audio from 360-degree video, we propose a novel framework OmniAudio, which leverages self-supervised pre-training using both spatial audio data (in FOA format) and large-scale non-spatial data. Furthermore, OmniAudio features a dual-branch framework that utilizes both panoramic and perspective video inputs to capture comprehensive local and global information from 360-degree videos. Experimental results demonstrate that OmniAudio achieves state-of-the-art performance across both objective and subjective metrics on Sphere360. Code and datasets are available at https://github.com/liuhuadai/OmniAudio. The project website is available at https://OmniAudio-360V2SA.github.io.

Bridge models in image restoration construct a diffusion process from degraded to clear images. However, existing methods typically require training a bridge model from scratch for each specific type of degradation, resulting in high computational costs and limited performance. This work aims to efficiently leverage pretrained generative priors within existing image restoration bridges to eliminate this requirement. The main challenge is that standard generative models are typically designed for a diffusion process that starts from pure noise, while restoration tasks begin with a low-quality image, resulting in a mismatch in the state distributions between the two processes. To address this challenge, we propose a transition equation that bridges two diffusion processes with the same endpoint distribution. Based on this, we introduce the IRBridge framework, which enables the direct utilization of generative models within image restoration bridges, offering a more flexible and adaptable approach to image restoration. Extensive experiments on six image restoration tasks demonstrate that IRBridge efficiently integrates generative priors, resulting in improved robustness and generalization performance. Code will be available at GitHub.

Large Language Models (LLMs) have exhibited exceptional performance in software engineering yet face challenges in adapting to continually evolving code knowledge, particularly the frequent updates of third-party library APIs. This limitation, rooted in the static pre-training datasets, often results in non-executable code or implementations with suboptimal safety and efficiency. To this end, we introduce CodeSync, a data engine to identify outdated code patterns and collect real-time code knowledge updates from Python third-party libraries. Building upon CodeSync, we develop CodeSyncBench, a comprehensive benchmark for assessing LLMs' ability to stay synchronized with code evolution, which covers real-world updates for 220 APIs from six Python libraries. Our benchmark offers 3,300 test cases spanning three evaluation tasks and an update-aware instruction tuning dataset of 2,200 training samples. Extensive experiments on 14 LLMs reveal that they struggle with dynamic code evolution, even with the support of advanced knowledge updating methods (e.g., DPO, ORPO, and SimPO). Our CodeSync lays a strong foundation for developing more effective and robust methods for real-time code knowledge updating in the future. The experimental code is available at: https://github.com/CGCL-codes/naturalcc/tree/main/examples/codesync.

Speech-driven 3D facial animation aims to create lifelike facial expressions that synchronize accurately with speech. Despite significant progress, many existing methods may focus on generating facial animation with a fixed emotional state, neglecting the diverse transformations of facial emotions under a given speech input. To solve this issue, we focus on exploring the refined alignment between speech representations and multiple domains in facial expression information. We aim to disentangle the spoken language and emotion facial priors from speech expressions, to guide the refinement of the facial vertices based on speech. To accomplish this objective, we propose ExpTalk, which first applies an Adaptive Disentanglement Variational Autoencoder (AD-VAE) to decouple facial expression aligned with spoken language and emotions of speech through contrastive learning. Then a Refined Alignment Diffusion (RAD) is employed to iteratively refine the decoupled facial expression priors through diffusion-based perturbations, producing facial animations that align with the emotional variations of the given speech. Extensive experiments prove the effectiveness of our ExpTalk by surpassing state-of-the-arts by a large margin.

Searching for images using text has limitations because language has difficulties in expressing certain abstract intentions, e.g. artistic styles are difficult to describe for non-experts. As for the image search image model, images can convey abstract intentions, but cannot express the specific purpose, so many of the current graph search works only have a single function, such as content search and style search. Our work aims to combine the strengths of both, merging the ability of text to express specific ideas with the ability of images to convey abstract concepts, thus achieving a better capture of the user's intentions. To this end, we propose CCSR, a multimodal conditional content-style joint retrieval model. Our model is the first to apply contrastive learning to conditional retrieval and introduces a novel Mixture-of-Expert models (MOE) system to enable collaboration between multiple expert systems. We adopt a novel prompt learning strategy that allows the model to adaptively select specific prompts, thereby enhancing its focus on the current task. In addition, to evaluate the joint content-style retrieval capability of our model, we present a new dataset, StyleCoco, containing rich content categories and style categories. The experimental results indicate that CCSR has achieved state-of-the-art performance in conditional style retrieval, content retrieval, and style-content retrieval. The dataset and code will be publicly available on https://mccsr.github.io/.

Text-guided diffusion models revolutionize audio generation by adapting source audio to specific text prompts. However, existing zero-shot audio editing methods such as DDIM inversion accumulate errors across diffusion steps, reducing the effectiveness. Moreover, existing editing methods struggle with conducting complex non-rigid music edits while maintaining content integrity and high fidelity. To address these challenges, we propose MelodyEdit, a novel zero-shot music editing system based on innovative Disentangled Inversion Control (DIC) technique, which comprises Harmonized Attention Control and Disentangled Inversion. Disentangled Inversion disentangles the diffusion process into triple branches to rectify the deviated path of the source branch caused by DDIM inversion. Harmonized Attention Control unifies the mutual self-attention control and the cross-attention control with an intermediate Harmonic Branch to progressively generate the desired harmonic and melodic information in the target music. We also introduce ZoME-Bench, a comprehensive music editing benchmark with 1,100 samples covering ten distinct editing categories. ZoME-Bench facilitates both zero-shot and instruction-based music editing tasks. Our method outperforms state-of-the-art inversion techniques in editing fidelity and content preservation.

Spatial audio playback defines immersive listening. However, objective evaluation methods for perceptual dimensions like sound field and sound image remain underdeveloped, hindered by the lack of fine-grained spatial audio datasets and the neglect of echoes and reverberation in diverse playback conditions. To address these challenges, we propose MESA, a multi-modal evaluation framework for spatial audio systems, and introduce PSA-MOS, a high-quality multi-scene spatial audio dataset. Specifically: 1) PSA-MOS provides 50 hours of high-quality spatial audio recordings spanning 6 playback scenarios and 7 device types, with detailed localization annotations and fine-grained MOS ratings across four perceptual dimensions. 2) We develop SAE-Encoder, a spatial audio encoder that captures both acoustic-spatial cues and fine-grained perceptual patterns. 3) MESA integrates visual scene context to enhance evaluation robustness through echo and reverberation modeling. Experimental results demonstrate that SAE-Encoder achieves superior performance in SELD tasks. With a two-stage training strategy, MESA exhibits strong correlation with human perceptual assessments, effectively guiding spatial audio quality optimization. The demos are available at https://david-pigeon.github.io/mesaDemo.

Hallucinations present a significant challenge in the development and evaluation of large language models (LLMs), directly affecting their reliability and accuracy. While notable advancements have been made in research on textual and visual hallucinations, there is still a lack of a comprehensive benchmark for evaluating auditory hallucinations in large audio language models (LALMs). To fill this gap, we introduce AHa-Bench, a systematic and comprehensive benchmark for audio hallucinations. Audio data, in particular, uniquely combines the multi-attribute complexity of visual data with the semantic richness of textual data, leading to auditory hallucinations that share characteristics with both visual and textual hallucinations. Based on the source of these hallucinations, AHa-Bench categorizes them into semantic hallucinations, acoustic hallucinations, and semantic-acoustic confusion hallucinations. In addition, we systematically evaluate seven open-source local perception language models (LALMs), demonstrating the challenges these models face in audio understanding, especially when it comes to jointly understanding semantic and acoustic information. Through the development of a comprehensive evaluation framework, AHa-Bench aims to enhance the robustness and stability of LALMs, fostering more reliable and nuanced audio understanding in LALMs. The benchmark dataset is available at \url{https://huggingface.co/datasets/ahabench/AHa-Bench}.

Industrial Anomaly Detection (IAD) is an indispensable quality control technology in modern production processes. Recently, on account of the outstanding visual comprehension and cross-domain knowledge transfer capabilities of multimodal large language models (MLLMs), existing studies have explored the application of MLLMs in the IAD domain and established some multimodal IAD datasets. However, although the latest datasets contain various fundamental IAD tasks, they formulate tasks in a general question-and-answer format lacking a rigorous reasoning process, and they are relatively limited in the diversity of scenarios, which restricts their reliability in practical applications. In this paper, we propose AnomalyCoT, a multimodal Chain-of-Thought (CoT) dataset for multi-scenario IAD tasks. It consists of 37,565 IAD samples with the CoT data and is defined by challenging composite IAD tasks. Meanwhile, the CoT data for each sample provides precise coordinates of anomaly regions, thereby improving visual comprehension of defects across different types. AnomalyCoT is constructed through a systematic pipeline and involves multiple manual operations. Based on AnomalyCoT, we conducted a comprehensive evaluation of various mainstream MLLMs and fine-tuned representative models in different ways. The final results show that Gemini-2.0-flash achieved the best performance in the direct evaluation with an accuracy rate of 59.6\%, while Llama 3.2-Vision achieves the best performance after LoRA fine-tuning with an accuracy rate of 94.0\%. Among all the fine-tuned models, the average accuracy improvement reaches 36.5\%, demonstrating the potential of integrating CoT datasets in future applications within the IAD field. The code and data are available at \url{https://github.com/Zhaolutuan/AnomalyCoT}.

The visual world around us constantly evolves, from real-time news and social media trends to global infrastructure changes visible through satellite imagery and augmented reality enhancements. However, Multimodal Large Language Models (MLLMs), which automate many tasks, struggle to stay current, limited by the cutoff dates in their fixed training datasets.To quantify this stagnation, we introduce LiveVQA, the first-of-its-kind dataset featuring 107,143 samples and 12 categories data specifically designed to support research in both seeking and updating with live visual knowledge.Drawing from recent news articles, video platforms, and academic publications in April 2024-May 2025, LiveVQA enables evaluation of how models handle latest visual information beyond their knowledge boundaries and how current methods help to update them. Our comprehensive benchmarking of 17 state-of-the-art MLLMs reveals significant performance gaps on content beyond knowledge cutoff, and tool-use or agentic visual seeking framework drastically gain an average of 327% improvement. Furthermore, we explore parameter-efficient fine-tuning methods to update MLLMs with new visual knowledge.We dive deeply to the critical balance between adapter capacity and model capability when updating MLLMs with new visual knowledge. All the experimental dataset and source code are publicly available at: https://livevqa.github.io.

Humans rely on multisensory integration to perceive spatial environments, where auditory cues enable sound source localization in three-dimensional space. Despite the critical role of spatial audio in immersive technologies such as VR/AR, most existing multimodal datasets provide only monaural audio, which limits the development of spatial audio generation and understanding. To address these challenges, we introduce MRSAudio, a large-scale multimodal spatial audio dataset designed to advance research in spatial audio understanding and generation. MRSAudio spans four distinct components: MRSLife, MRSSpeech, MRSMusic, and MRSSing, covering diverse real-world scenarios. The dataset includes synchronized binaural and ambisonic audio, exocentric and egocentric video, motion trajectories, and fine-grained annotations such as transcripts, phoneme boundaries, lyrics, scores, and prompts.To demonstrate the utility and versatility of MRSAudio, we establish five foundational tasks: audio spatialization, and spatial text to speech, spatial singing voice synthesis, spatial music generation and sound event localization and detection. Results show that MRSAudio enables high-quality spatial modeling and supports a broad range of spatial audio research.Demos and dataset access are available at https://mrsaudio.github.io.

With the continuous development of large language models and reasoning chain technologies, the potential of deep reasoning based on reinforcement learning has shown remarkable promise in multi-task scenarios. However, existing unified models have yet to achieve end-to-end integration in image generation and understanding tasks, limiting the model’s self-reflection ability and the realization of cross-modal reasoning chains. To address this, we propose Vinic, a novel framework designed to enable interleaved image generation and understanding through deep reasoning capabilities. We leverage a small amount of multimodal chain-of-thought (MCoT) data for cold-start and employ reinforcement learning to guide the integration of image generation and understanding tasks. Additionally, we introduce a momentum-based reward function, which dynamically adjusts the reward distribution by considering historical improvements, ensuring the stability of the model across multiple generations. Experimental results demonstrate that integrating MCoT can achieve a +22% improvement over the base model on Geneval, effectively enhancing both image generation quality and instruction alignment capabilities.

While end-to-end video-to-audio generation has greatly improved, producing high-fidelity audio that authentically captures the nuances of visual content remains challenging. Like professionals in the creative industries, this generation requires sophisticated reasoning about items such as visual dynamics, acoustic environments, and temporal relationships. We present **ThinkSound**, a novel framework that leverages Chain-of-Thought (CoT) reasoning to enable stepwise, interactive audio generation and editing for videos. Our approach decomposes the process into three complementary stages: foundational foley generation that creates semantically coherent soundscapes, interactive object-centric refinement through precise user interactions, and targeted editing guided by natural language instructions. At each stage, a multimodal large language model generates contextually aligned CoT reasoning that guides a unified audio foundation model. Furthermore, we introduce **AudioCoT**, a comprehensive dataset with structured reasoning annotations that establishes connections between visual content, textual descriptions, and sound synthesis. Experiments demonstrate that ThinkSound achieves state-of-the-art performance in video-to-audio generation across both audio metrics and CoT metrics, and excels in the out-of-distribution Movie Gen Audio benchmark. The project page is available at https://ThinkSound-Project.github.io.

Humans can intuitively compose and arrange scenes in the 3D space for photography. However, can advanced AI image generators plan scenes with similar 3D spatial awareness when creating images from text or image prompts? We present GenSpace, a novel benchmark and evaluation pipeline to comprehensively assess the spatial awareness of current image generation models. Furthermore, standard evaluations using general Vision-Language Models (VLMs) frequently fail to capture the detailed spatial errors. To handle this challenge, we propose a specialized evaluation pipeline and metric, which reconstructs 3D scene geometry using multiple visual foundation models and provides a more accurate and human-aligned metric of spatial faithfulness. Our findings show that while AI models create visually appealing images and can follow general instructions, they struggle with specific 3D details like object placement, relationships, and measurements. We summarize three core limitations in the spatial perception of current state-of-the-art image generation models: 1) Object Perspective Understanding, 2) Egocentric-Allocentric Transformation, and 3) Metric Measurement Adherence, highlighting possible directions for improving spatial intelligence in image generation.

This work presents Orient Anything V2, an enhanced foundation model for unified understanding of object 3D orientation and rotation from single or paired images. Building upon Orient Anything V1, which defines orientation via a single unique front face, V2 extends this capability to handle objects with diverse rotational symmetries and directly estimate relative rotations. These improvements are enabled by four key innovations: 1) Scalable 3D assets synthesized by generative models, ensuring broad category coverage and balanced data distribution; 2) An efficient, model-in-the-loop annotation system that robustly identifies 0 to N valid front faces for each object; 3) A symmetry-aware, periodic distribution fitting objective that captures all plausible front-facing orientations, effectively modeling object rotational symmetry; 4) A multi-frame architecture that directly predicts relative object rotations. Extensive experiments show that Orient Anything V2 achieves state-of-the-art zero-shot performance on orientation estimation, 6DoF pose estimation, and object symmetry recognition across 11 widely used benchmarks. The model demonstrates strong generalization, significantly broadening the applicability of orientation estimation in diverse downstream tasks.

Autoregressive models (ARMs) show strong capabilities in many domains but face challenges with planning and complex reasoning due to their sequential generation. Masked diffusion models (MDMs) address these issues by enabling controllable, any-order, and parallel generation but encounter training difficulties as token prediction complexity varies with unmasked token positions. This work identifies two key characteristics of efficient MDM sampling paths: prioritizing tokens near unmasked ones and generating subsequence earlier in reasoning. We propose the Simple Path Masked Diffusion Model (SPMDM), which partitions sequences into fixed-length, non-overlapping subsequences and applies varying noise scales to learn token-level and cross-subsequence dependencies. Experiments on synthetic data and tasks like Countdown and Sudoku show SPMDM captures structural rules effectively, significantly outperforming existing MDMs and ARMs, with competitive results on broader reasoning benchmarks.

Large language models (LLMs) are increasingly leveraged as foundational backbones in the development of advanced recommender systems, offering enhanced capabilities through their extensive knowledge and reasoning.Existing llm-based recommender systems (RSs) often face challenges due to the significant differences between the linguistic semantics of pre-trained LLMs and the collaborative semantics essential for RSs.These systems use pre-trained linguistic semantics but learn collaborative semantics from scratch via the llm-Backbone.However, LLMs are not designed for recommendations, leading to inefficient collaborative learning, weak result correlations, and poor integration of traditional RS features.To address these challenges, we propose EAGER-LLM, a decoder-only llm-based generative recommendation framework that integrates endogenous and exogenous behavioral and semantic information in a non-intrusive manner.Specifically, we propose 1) dual-source knowledge-rich item indices that integrates indexing sequences * Both authors contributed equally to this research.

Recently, audio–visual speech recognition (AVSR) has attracted increasing attention. However, most existing works simplify the complex challenges in real-world applications and only focus on scenarios with two speakers and perfectly aligned audio-video clips. In this work, we study the effect of speaker number and modal misalignment in the AVSR task, and propose an end-to-end AVSR framework under a more realistic condition. Specifically, we propose a speaker-number-aware mixture-of-experts (SA-MoE) mechanism to explicitly model the characteristic difference in scenarios with different speaker numbers, and a cross-modal realignment (CMR) module for robust handling of asynchronous inputs. We also use the underlying difficulty difference and introduce a new training strategy named challenge-based curriculum learning (CBCL), which forces the model to focus on difficult, challenging data instead of simple data to improve efficiency.

Large language models (LLMs) have exhibited remarkable capabilities across a variety of domains and tasks, challenging our understanding of learning and cognition. Despite the recent success, current LLMs are not capable of processing complex audio information or conducting spoken conversations (like Siri or Alexa). In this work, we propose a multi-modal AI system named AudioGPT, which complements LLMs (i.e., ChatGPT) with 1) foundation models to process complex audio information and solve numerous understanding and generation tasks; and 2) the input/output interface (ASR, TTS) to support spoken dialogue. With an increasing demand to evaluate multi-modal LLMs of human intention understanding and cooperation with foundation models, we outline the principles and processes and test AudioGPT in terms of consistency, capability, and robustness. Experimental results demonstrate the capabilities of AudioGPT in solving 16 AI tasks with speech, music, sound, and talking head understanding and generation in multi-round dialogues, which empower humans to create rich and diverse audio content with unprecedented ease. Code can be found in https://github.com/AIGC-Audio/AudioGPT

Style transfer for out-of-domain (OOD) singing voice synthesis (SVS) focuses on generating high-quality singing voices with unseen styles (such as timbre, emotion, pronunciation, and articulation skills) derived from reference singing voice samples. However, the endeavor to model the intricate nuances of singing voice styles is an arduous task, as singing voices possess a remarkable degree of expressiveness. Moreover, existing SVS methods encounter a decline in the quality of synthesized singing voices in OOD scenarios, as they rest upon the assumption that the target vocal attributes are discernible during the training phase. To overcome these challenges, we propose StyleSinger, the first singing voice synthesis model for zero-shot style transfer of out-of-domain reference singing voice samples. StyleSinger incorporates two critical approaches for enhanced effectiveness: 1) the Residual Style Adaptor (RSA) which employs a residual quantization module to capture diverse style characteristics in singing voices, and 2) the Uncertainty Modeling Layer Normalization (UMLN) to perturb the style attributes within the content representation during the training phase and thus improve the model generalization. Our extensive evaluations in zero-shot style transfer undeniably establish that StyleSinger outperforms baseline models in both audio quality and similarity to the reference singing voice samples. Access to singing voice samples can be found at https://stylesinger.github.io/.

The development of multimodal models has significantly advanced multimodal sentiment analysis and emotion recognition.However, in real-world applications, the presence of various missing modality cases often leads to a degradation in the model's performance.In this work, we propose a novel multimodal Transformer framework using prompt learning to address the issue of missing modalities.Our method introduces three types of prompts: generative prompts, missing-signal prompts, and missingtype prompts.These prompts enable the generation of missing modality features and facilitate the learning of intra-and inter-modality information.Through prompt learning, we achieve a substantial reduction in the number of trainable parameters.Our proposed method outperforms other methods significantly across all evaluation metrics.Extensive experiments and ablation studies are conducted to demonstrate the effectiveness and robustness of our method, showcasing its ability to effectively handle missing modalities.Codes are available at https://github.com/zrguo/MPLMM.

Zhiqing Hong, Rongjie Huang, Xize Cheng, Yongqi Wang, Ruiqi Li, Fuming You, Zhou Zhao, Zhimeng Zhang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Rongjie Huang, Chunlei Zhang, Yongqi Wang, Dongchao Yang, Jinchuan Tian, Zhenhui Ye, Luping Liu, Zehan Wang, Ziyue Jiang, Xuankai Chang, Jiatong Shi, Chao Weng, Zhou Zhao, Dong Yu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Zero-shot text-to-speech (TTS) has gained significant attention due to its powerful voice cloning capabilities, requiring only a few seconds of unseen speaker voice prompts.However, all previous work has been developed for cloud-based systems.Taking autoregressive models as an example, although these approaches achieve high-fidelity voice cloning, they fall short in terms of inference speed, model size, and robustness.Therefore, we propose MobileSpeech, which is a fast, lightweight, and robust zero-shot text-tospeech system based on mobile devices for the first time.Specifically: 1) leveraging discrete codec, we design a parallel speech mask decoder module called SMD, which incorporates hierarchical information from the speech codec and weight mechanisms across different codec layers during the generation process.Moreover, to bridge the gap between text and speech, we introduce a high-level probabilistic mask that simulates the progression of information flow from less to more during speech generation.2) For speaker prompts, we extract fine-grained prompt duration from the prompt speech and incorporate text, prompt speech by cross attention in SMD.We demonstrate the effectiveness of MobileSpeech on multilingual datasets at different levels, achieving state-ofthe-art results in terms of generating speed and speech quality.MobileSpeech achieves RTF of 0.09 on a single A100 GPU and we have successfully deployed MobileSpeech on mobile devices.Audio samples are available at https://mobilespeech.github.io/ .

Transformer-based methods have gone mainstream in multimodal sequential learning.The intra and inter modality interactions are captured by the query-key associations of multihead attention.In this way, the calculated multimodal contexts (attentional results) are expected to be relevant to the query modality.However, in existing literature, the alignment degree between different calculated attentional results of the same query are under-explored.Based on this concern, we propose a new constrained scheme called Multimodal Contextual Contrast (MCC), which could align the multiple attentional results from both local and global perspectives, making the information capture more efficient.Concretely, the calculated attentional results of different modalities are mapped into a common feature space, those attentional vectors with the same query are considered as a positive group and the remaining sets are negative.From local perspective, we sample the negative groups for a positive group by randomly changing the sequential step of one specific context and keeping the other stay the same.From coarse global perspective, we divide all the contextual groups into two sets (i.e., aligned and unaligned), making the total score of aligned group relatively large.We extend the vectorial inner product operation for more input and calculate the aligned score for each multimodal group.Considering that the computational complexity scales exponentially to the number of modalities, we adopt stochastic expectation approximation (SEA) for the real process.The extensive experimental results on several tasks reveal the effectiveness of our contributions.

Songju Lei, Xize Cheng, Mengjiao Lyu, Jianqiao Hu, Jintao Tan, Runlin Liu, Lingyu Xiong, Tao Jin, Xiandong Li, Zhou Zhao. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Note-level Automatic Singing Voice Transcription (AST) converts singing recordings into note sequences, facilitating the automatic annotation of singing datasets for Singing Voice Synthesis (SVS) applications.Current AST methods, however, struggle with accuracy and robustness when used for practical annotation.This paper presents ROSVOT, the first robust AST model that serves SVS, incorporating a multi-scale framework that effectively captures coarse-grained note information and ensures fine-grained frame-level segmentation, coupled with an attention-based pitch decoder for reliable pitch prediction.We also established a comprehensive annotation-and-training pipeline for SVS to test the model in realworld settings.Experimental findings reveal that ROSVOT achieves state-of-the-art transcription accuracy with either clean or noisy inputs.Moreover, when trained on enlarged, automatically annotated datasets, the SVS model outperforms its baseline, affirming the capability for practical application.Audio samples are available at https://rosvot.github.io.Codes can be found at https://github.com/RickyL- 2000/ROSVOT.

Qian Yang, Jin Xu, Wenrui Liu, Yunfei Chu, Ziyue Jiang, Xiaohuan Zhou, Yichong Leng, Yuanjun Lv, Zhou Zhao, Chang Zhou, Jingren Zhou. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Recent advancements in single image driven 3D content generation have been propelled by leveraging prior knowledge from pretrained 2D diffusion models. However, the 3D content generated by existing methods often exhibits distorted outline shapes and inadequate details. To solve this problem, we propose a novel framework called Mask-enhanced Progressive Outline-to-Detail optimization (aka. MPOD123), which consists of two stages. Specifically, in the first stage, MPOD123 utilizes the pretrained view-conditioned diffusion model to guide the outline shape optimization of the 3D content. Given certain viewpoint, we estimate outline shape priors in the form of 2D mask from the 3D content by leveraging opacity calculation. In the second stage, MPOD123 incorporates Detail Appearance Inpainting (DAI) to guide the refinement on local geometry and texture with the shape priors. The essence of DAI lies in the Mask Rectified Cross-Attention (MRCA), which can be conveniently plugged in the stable diffusion model. The MRCA module utilizes the mask to rectify the attention map from each cross-attention layer. Accompanied with this new module, DAI is capable of guiding the detail refinement of the 3D content, while better preserves the outline shape. To assess the applicability in practical scenarios, we contribute a new dataset modeled on real-world e-commerce environments. Extensive quantitative and qualitative experiments on this dataset and open benchmarks demonstrate the effectiveness of MPOD123 over the state-of-the-arts.

Zero-shot singing voice synthesis (SVS) with style transfer and style control aims to generate high-quality singing voices with unseen timbres and styles (including singing method, emotion, rhythm, technique, and pronunciation) from audio and text prompts.However, the multifaceted nature of singing styles poses a significant challenge for effective modeling, transfer, and control.Furthermore, current SVS models often fail to generate singing voices rich in stylistic nuances for unseen singers.To address these challenges, we introduce TCSinger, the first zero-shot SVS model for style transfer across cross-lingual speech and singing styles, along with multi-level style control.Specifically, TCSinger proposes three primary modules: 1) the clustering style encoder employs a clustering vector quantization model to stably condense style information into a compact latent space; 2) the Style and Duration Language Model (S&D-LM) concurrently predicts style information and phoneme duration, which benefits both; 3) the style adaptive decoder uses a novel mel-style adaptive normalization method to generate singing voices with enhanced details.Experimental results show that TCSinger outperforms all baseline models in synthesis quality, singer similarity, and style controllability across various tasks, including zero-shot style transfer, multi-level style control, crosslingual style transfer, and speech-to-singing style transfer.Singing voice samples can be accessed at https://tcsinger.github.io/.

Recently, there has been a growing interest in the field of controllable Text-to-Speech (TTS). While previous studies have relied on users providing specific style factor values based on acoustic knowledge or selecting reference speeches that meet certain requirements, generating speech solely from natural text prompts has emerged as a new challenge for researchers. This challenge arises due to the scarcity of high-quality speech datasets with natural text style prompt and the absence of advanced text-controllable TTS models. In light of this, 1) we propose TextrolSpeech, which is the first large-scale speech emotion dataset annotated with rich text attributes. The dataset comprises 236,203 pairs of style prompt in natural text descriptions with five style factors and corresponding speech samples. Through iterative experimentation, we introduce a multi-stage prompt programming approach that effectively utilizes the GPT model for generating natural style descriptions in large volumes. 2) Furthermore, to address the need for generating audio with greater style diversity, we propose an efficient architecture called Salle. This architecture treats text controllable TTS as a language model task, utilizing audio codec codes as an intermediate representation to replace the conventional mel-spectrogram. Finally, we successfully demonstrate the ability of the proposed model by showing a comparable performance in the controllable TTS task. Audio samples are available on the demo page https://sall-e.github.io/.

The primary objective of simultaneous machine translation (SiMT) is to minimize latency while preserving the quality of the final translation. Drawing inspiration from CPU branch prediction techniques, we propose incorporating branch prediction techniques in SiMT tasks to reduce translation latency. Specifically, we utilize a language model as a branch predictor to predict potential branch directions, namely, future source words. Subsequently, we utilize the predicted source words to decode the output in advance. When the actual source word deviates from the predicted source word, we use the real source word to decode the output again, replacing the predicted output. To further reduce computational costs, we share the parameters of the encoder and the branch predictor, and utilize a pre-trained language model for initialization. Our proposed method can be seamlessly integrated with any SiMT model. Extensive experimental results demonstrate that our approach can improve translation quality and latency at the same time. Our code is available at https://github.com/YinAoXiong/simt_branch_predictor.

Zero-shot text-to-speech (TTS) aims to synthesize voices with unseen speech prompts, which significantly reduces the data and computation requirements for voice cloning by skipping the fine-tuning process. However, the prompting mechanisms of zero-shot TTS still face challenges in the following aspects: 1) previous works of zero-shot TTS are typically trained with single-sentence prompts, which significantly restricts their performance when the data is relatively sufficient during the inference stage. 2) The prosodic information in prompts is highly coupled with timbre, making it untransferable to each other. This paper introduces Mega-TTS 2, a generic prompting mechanism for zero-shot TTS, to tackle the aforementioned challenges. Specifically, we design a powerful acoustic autoencoder that separately encodes the prosody and timbre information into the compressed latent space while providing high-quality reconstructions. Then, we propose a multi-reference timbre encoder and a prosody latent language model (P-LLM) to extract useful information from multi-sentence prompts. We further leverage the probabilities derived from multiple P-LLM outputs to produce transferable and controllable prosody. Experimental results demonstrate that Mega-TTS 2 could not only synthesize identity-preserving speech with a short prompt of an unseen speaker from arbitrary sources but consistently outperform the fine-tuning method when the volume of data ranges from 10 seconds to 5 minutes. Furthermore, our method enables to transfer various speaking styles to the target timbre in a fine-grained and controlled manner. Audio samples can be found in https://boostprompt.github.io/boostprompt/.

One-shot 3D talking portrait generation aims to reconstruct a 3D avatar from an unseen image, and then animate it with a reference video or audio to generate a talking portrait video. The existing methods fail to simultaneously achieve the goals of accurate 3D avatar reconstruction and stable talking face animation. Besides, while the existing works mainly focus on synthesizing the head part, it is also vital to generate natural torso and background segments to obtain a realistic talking portrait video. To address these limitations, we present Real3D-Potrait, a framework that (1) improves the one-shot 3D reconstruction power with a large image-to-plane model that distills 3D prior knowledge from a 3D face generative model; (2) facilitates accurate motion-conditioned animation with an efficient motion adapter; (3) synthesizes realistic video with natural torso movement and switchable background using a head-torso-background super-resolution model; and (4) supports one-shot audio-driven talking face generation with a generalizable audio-to-motion model. Extensive experiments show that Real3D-Portrait generalizes well to unseen identities and generates more realistic talking portrait videos compared to previous methods. Video samples are available at https://real3dportrait.github.io.

Unified multi-model representation spaces are the foundation of multimodal understanding and generation. However, the billions of model parameters and catastrophic forgetting problems make it challenging to further enhance pre-trained unified spaces. In this work, we propose FreeBind, an idea that treats multimodal representation spaces as basic units, and freely augments pre-trained unified space by integrating knowledge from extra expert spaces via “space bonds". Specifically, we introduce two kinds of basic space bonds: 1) Space Displacement Bond and 2) Space Combination Bond. Based on these basic bonds, we design Complex Sequential & Parallel Bonds to effectively integrate multiple spaces simultaneously. Benefiting from the modularization concept, we further propose a coarse-to-fine customized inference strategy to flexibly adjust the enhanced unified space for different purposes. Experimentally, we bind ImageBind with extra image-text and audio-text expert spaces, resulting in three main variants: ImageBind++, InternVL_IB, and InternVL_IB++. These resulting spaces outperform ImageBind on 5 audio-image-text downstream tasks across 9 datasets. Moreover, via customized inference, it even surpasses the advanced audio-text and image-text expert spaces. Our code and checkpoints are released at https://github.com/zehanwang01/FreeBind

Instruction-guided speech editing aims to follow the user’s natural language instruction to manipulate the semantic and acoustic attributes of a speech. In this work, we construct triplet paired data (instruction, input speech, output speech) to alleviate data scarcity and train a multi-task large language model named InstructSpeech. To mitigate the challenges of accurately executing user’s instructions, we 1) introduce the learned task embeddings with a fine-tuned Flan-T5-XL to guide the generation process towards the correct generative task; 2) include an extensive and diverse set of speech editing and processing tasks to enhance model capabilities; 3) investigate chain-of-thought reasoning for free-form semantic content editing; and 4) propose a hierarchical adapter that effectively updates a small portion of parameters for generalization to new tasks. To assess instruction speech editing in greater depth, we introduce a benchmark evaluation with contrastive instruction-speech pre-training (CISP) to test the speech quality and instruction-speech alignment faithfulness. Experimental results demonstrate that InstructSpeech achieves state-of-the-art results in eleven tasks, for the first time unlocking the ability to edit speech’s acoustic and semantic attributes following a user’s instruction. Audio samples are available at https://InstructSpeech.github.io

We tackle the common challenge of inter-concept visual confusion in compositional concept generation using text-guided diffusion models (TGDMs). It becomes even more pronounced in the generation of customized concepts, due to the scarcity of user-provided concept visual examples. By revisiting the two major stages leading to the success of TGDMs---1) contrastive image-language pre-training (CLIP) for text encoder that encodes visual semantics, and 2) training TGDM that decodes the textual embeddings into pixels---we point that existing customized generation methods only focus on fine-tuning the second stage while overlooking the first one. To this end, we propose a simple yet effective solution called CLIF: contrastive image-language fine-tuning. Specifically, given a few samples of customized concepts, we obtain non-confusing textual embeddings of a concept by fine-tuning CLIP via contrasting a concept and the over-segmented visual regions of other concepts. Experimental results demonstrate the effectiveness of CLIF in preventing the confusion of multi-customized concept generation. Project page: https://clif-official.github.io/clif.

Audio generation is a major branch of generative AI research. Compared with prior works in this area that are commonly task-specific with heavy domain knowledge, this paper advocates building universal audio generation models that can handle various tasks in a unified manner. As recent research on large language models (LLMs) has demonstrated their strong ability to handle multiple tasks, this work presents UniAudio, an LLM-based audio generation model that supports a wide range of audio generation tasks. Based on various input conditions, such as phoneme, text description, or audio itself, UniAudio can generate speech, sound, music, and singing voice. The proposed UniAudio is built with 100k hours of multi-source open-available audio data and is scaled to 1B parameters. The audio tokenization method and language model architecture are also specifically designed for both performance and efficiency. Experimentally, UniAuido supports 11 audio generation tasks and achieves competitive results on all tasks consistently. We also show that UniAudio can support new tasks seamlessly via simple fine-tuning.

Personalized recommendation serves as a ubiquitous channel for users to discover information tailored to their interests. However, traditional recommendation models primarily rely on unique IDs and categorical features for user-item matching, potentially overlooking the nuanced essence of raw item contents across multiple modalities such as text, image, audio, and video. This underutilization of multimodal data poses a limitation to recommender systems, especially in multimedia services like news, music, and short-video platforms. The recent advancements in large multimodal models offer new opportunities and challenges in developing content-aware recommender systems. This survey seeks to provide a comprehensive exploration of the latest advancements and future trajectories in multimodal pretraining, adaptation, and generation techniques, as well as their applications in enhancing recommender systems. Furthermore, we discuss current open challenges and opportunities for future research in this dynamic domain. We believe that this survey, alongside the curated resources, will provide valuable insights to inspire further advancements in this evolving landscape.

Generative retrieval has recently emerged as a promising approach to sequential recommendation, framing candidate item retrieval as an autoregressive sequence generation problem. However, existing generative methods typically focus solely on either behavioral or semantic aspects of item information, neglecting their complementary nature and thus resulting in limited effectiveness. To address this limitation, we introduce EAGER, a novel generative recommendation framework that seamlessly integrates both behavioral and semantic information. Specifically, we identify three key challenges in combining these two types of information: a unified generative architecture capable of handling two feature types, ensuring sufficient and independent learning for each type, and fostering subtle interactions that enhance collaborative information utilization. To achieve these goals, we propose (1) a two-stream generation architecture leveraging a shared encoder and two separate decoders to decode behavior tokens and semantic tokens with a confidence-based ranking strategy; (2) a global contrastive task with summary tokens to achieve discriminative decoding for each type of information; and (3) a semantic-guided transfer task designed to implicitly promote cross-interactions through reconstruction and estimation objectives. We validate the effectiveness of EAGER on four public benchmarks, demonstrating its superior performance compared to existing methods. Our source code will be publicly available on PapersWithCode.com.

Hypothesis inference, a sophisticated cognitive process that allows humans to construct plausible explanations for incomplete observations, is paramount to our ability to make sense of the world around us. Despite the universality of this skill, it remains under-explored within the context of multi-modal AI, which necessitates analyzing observation, recalling information in the mind, and generating explanations. In this work, we propose the Cross-modal Observation hypothesIs iNference task (COIN). Given a textual description of a partially observed event, COIN strives to recall the most probable event from the visual mind (video pool), and infer the subsequent action flow connecting the visual mind event and the observed textural event. To advance the development of this field, we propose a large-scale text-video dataset, Tex-COIN, that contains 39,796 meticulously annotated hypothesis inference examples and auxiliary commonsense knowledge (appearance, clothing, action, etc.) for key video characters. Based on the proposed Tex-COIN dataset, we design a strong baseline, COINNet, which features two perspectives: 1) aligning temporally displaced textual observations with target videos via transformer-based multi-task learning, and 2) inferring the action flow with non-parametric graph-based inference grounded in graph theory. Extensive experiments on the Tex-COIN dataset validate the effectiveness of our COINNet by significantly outperforming the state-of-the-arts.

In the field of machine learning, continual learning is a crucial concept that allows models to adapt to non-stationary data distributions. However, most of the existing works focus on uni-modal settings and ignore the multi-modal data. In this paper, to enable neural networks better understand diverse modalities in real-world scenario, we investigate continual learning for two typical vision-language applications, i.e. retrieval and grounding. Instead of conventional exemplar-based methods, we leverage the pre-trained transformer model (e.g. CLIP/GLIP) and the prompt technique to tackle this problem. Under this scheme, we identify two critical limitations in existing methods: (1) Unfamiliarity across tasks, which prevents task-specific prompts from achieving forward propagation; and (2) Heterogeneity between modalities, which makes it difficult to guarantee a consistent optimization direction for prompts of different modalities. To overcome these constraints, we design Historical Prompt Calibration that includes two objectives to calibrate prompts. First, the intra-modal relevance estimation helps encode sufficient task-specific information for prompts, with the help a relevance estimator developed for recognizing task relevance. Second, the inter-modal consistency alignment enhances the agreement of the two modality-specific prompts in the current task by contrasting them with the prompts from previous tasks. We evaluate the superiority of our strategy over state-of-the arts methods by four vision-language applications, including two retrieval tasks (i.e. image- and video-text retrieval) and two grounding tasks (i.e. referring expression comprehension and segmentation).

In the burgeoning field of Audio-Visual Speech Recognition (AVSR), extant research has predominantly concentrated on the training paradigms tailored for high-quality resources. However, owing to the challenges inherent in real-world data collection, audio-visual data are frequently affected by modality-distortion, which encompasses audio-visual asynchrony, video noise and audio noise. The recognition accuracy of existing AVSR method is significantly compromised when multiple modality-distortion coexist in low-resource data. In light of the above challenges, we propose PCD: cluster-Prompt with Contrastive Decomposition, a robust framework for modality-distortion speech recognition, specifically devised to transpose the pre-trained knowledge from high-resource domain to the targeted domain by leveraging contrast-augmented prompts. In contrast to previous studies, we take into consideration the possibility of various types of distortion in both the audio and visual modalities. Concretely, we design bespoke prompts to delineate each modality-distortion, guiding the model to achieve speech recognition applicable to various distortion scenarios with quite few learnable parameters. To materialize the prompt mechanism, we employ multiple cluster-based strategies that better suits the pre-trained audio-visual model. Additionally, we design a contrastive decomposition mechanism to restrict the explicit relationships among various modality conditions, given their shared task knowledge and disparate modality priors. Extensive results on LRS2 dataset demonstrate that PCD achieves state-of-the-art performance for audio-visual speech recognition under the constraints of distorted resources. Code is available at https://github.com/ballooncatt/PCD.

Voice large language models (LLMs) cast voice synthesis as a language modeling task in a discrete space, and have demonstrated significant progress to date. Despite the recent success, the current development of voice LLMs in low-resource applications is hampered by data scarcity and high computational cost. In this work, we propose VoiceTuner, with a self-supervised pre-training and efficient fine-tuning approach for low-resource voice generation. Specifically, 1) to mitigate data scarcity, we leverage large-scale unlabeled dataset and pre-train VoiceTuner-SSL without pre-defined applications, which can be fine-tuned in downstream tasks; 2) to further reduce the high training cost in complete fine-tuning, we introduce a multiscale transformer adapter to effectively update only around 1% parameters as a plug-and-play module. Experimental results demonstrate that VoiceTuner-SSL presents strong acoustic continuations, and VoiceTuner achieves state-of-the-art results in rich-resource TTS evaluation compared with competitive baseline models. Low-resource (1h, 10h, 30h) downstream applications including zero-shot TTS, instruction TTS, and singing voice synthesis present VoiceTuner's superior audio quality and style similarity with reduced data requirement and computational cost. Audio samples are available at https://VoiceTuner.github.io

Dynamic sequential recommendation (DSR) can generate model parameters based on user behavior to improve the personalization of sequential recommendation under various user preferences. However, it faces the challenges of large parameter search space and sparse and noisy user-item interactions, which reduces the applicability of the generated model parameters. The Semantic Codebook Learning for Dynamic Recommendation Models (SOLID) framework presents a significant advancement in DSR by effectively tackling these challenges. By transforming item sequences into semantic sequences and employing a dual parameter model, SOLID compresses the parameter generation search space and leverages homogeneity within the recommendation system. The introduction of the semantic metacode and semantic codebook, which stores disentangled item representations, ensures robust and accurate parameter generation. Extensive experiments demonstrates that SOLID consistently outperforms existing DSR, delivering more accurate, stable, and robust recommendations.

Research on Multi-modal Large Language Models (MLLMs) towards the multi-image cross-modal instruction has received increasing attention and made significant progress, particularly in scenarios involving closely resembling images ( e.g., change captioning). Existing MLLMs typically follow a two-step process in their pipelines: first, extracting visual tokens independently for each input image, and then aligning these visual tokens from different images with the Large Language Model (LLM) in its textual feature space. However, the independent extraction of visual tokens for each image may result in different semantics being prioritized for different images in the first step, leading to a lack of preservation of linking information among images for subsequent LLM analysis. This issue becomes more serious in scenarios where significant variations exist among the images (e.g., visual storytelling). To address this challenge, we introduce Semantic Alignment for Multi-modal large language models (SAM). By involving the bidirectional semantic guidance between different images in the visual-token extraction process, SAM aims to enhance the preservation of linking information for coherent analysis and align the semantics of different images before feeding them into LLM. As the test bed, we propose a large-scale dataset named MmLINK consisting of 69K samples. Different from most existing datasets for MLLMs fine-tuning, our MmLINK dataset comprises multi-modal instructions with significantly diverse images. Extensive experiments on the group captioning task and the storytelling task prove the effectiveness of our SAM model, surpassing the state-of-the-art methods by a large margin (+37% for group captioning and +22% for storytelling on CIDEr score). Project page: https://mccartney01.github.io/SAM.

Research on continual learning in multi-modal tasks has been receiving increasing attention. However, most existing work overlooks the explicit cross-modal and cross-task interactions. In this paper, we innovatively propose the Low-rank Prompt Interaction (LPI) to address this general problem of multi-modal understanding, which considers both cross-modal and cross-task interactions. Specifically, as for the former, we employ multi-modal correlation modules for corresponding Transformer layers. Considering that the training parameters scale to the number of layers and tasks, we propose low-rank interaction-augmented decomposition to avoid memory explosion while enhancing the cross-modal association through sharing and separating common-specific low-rank factors. In addition, due to the multi-modal semantic differences carried by the low-rank initialization, we adopt hierarchical low-rank contrastive learning to ensure training robustness. As for the latter, we initially employ a visual analysis and identify that different tasks have clear distinctions in proximity. Therefore, we introduce explicit task contrastive constraints in the prompt learning process based on task semantic distances. Experiments on two retrieval tasks show performance improvements with the introduction of a minimal number of parameters, demonstrating the effectiveness of our method. Code is available at https://github.com/Kelvin-ywc/LPI.

Talking Face Generation (TFG) reconstructs facial motions concerning lips given speech input, which aims to generate highquality, synchronized, and lip-readable videos. Previous efforts have achieved success in generating quality and synchronization, and recently, there has been an increasing focus on the importance of intelligibility. Despite these efforts, there remains a challenge in achieving a balance among quality, synchronization, and intelligibility, often resulting in trade-offs that compromise one aspect in favor of another. In light of this, we propose SyncTalklip, a novel dual-tower framework designed to overcome the challenges of synchronization while improving lip-reading performance. To enhance the performance of SyncTalklip in both synchronization and intelligibility, we design AV-SyncNet, a pre-trained multi-task model, aiming to achieve a dual-focus on synchronization and intelligibility. Moreover, we propose a novel cross-modal contrastive learning bringing audio and video closer to enhance synchronization. Experimental results demonstrate that SyncTalklip achieves state-of-the-art performance in quality, intelligibility, and synchronization. Furthermore, extensive experiments have demonstrated our model's generalizability across domains. The code and demo is available at https://sync-talklip.github.io.

Yongqi Wang, Ruofan Hu, Rongjie Huang, Zhiqing Hong, Ruiqi Li, Wenrui Liu, Fuming You, Tao Jin, Zhou Zhao. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Multimodal learning has developed very fast in recent years. However, during the multimodal training process, the model tends to rely on only one modality based on which it could learn faster, thus leading to inadequate use of other modalities. Existing methods to balance the training process always have some limitations on the loss functions, optimizers and the number of modalities and only consider modulating the magnitude of the gradients while ignoring the directions of the gradients. To solve these problems, in this paper, we present a novel method to balance multimodal learning with **C**lassifier-**G**uided **G**radient **M**odulation (CGGM), considering both the magnitude and directions of the gradients. We conduct extensive experiments on four multimodal datasets: UPMC-Food 101, CMU-MOSI, IEMOCAP and BraTS 2021, covering classification, regression and segmentation tasks. The results show that CGGM outperforms all the baselines and other state-of-the-art methods consistently, demonstrating its effectiveness and versatility. Our code is available at https://github.com/zrguo/CGGM.

Recent advancements in 3D Large Language Models (LLMs) have demonstrated promising capabilities for 3D scene understanding. However, previous methods exhibit deficiencies in general referencing and grounding capabilities for intricate scene comprehension. In this paper, we introduce the use of object identifiers and object-centric representations to interact with scenes at the object level. Specifically, we decompose the input 3D scene into a set of object proposals, each assigned a unique identifier token, which enables efficient object referencing and grounding during user-assistant interactions. Given the scarcity of scene-language data, we model the scene embeddings as a sequence of explicit object-level embeddings, derived from semantic-rich 2D or 3D representations. By employing object identifiers, we transform diverse 3D scene-language tasks into a unified question-answering format, facilitating joint training without the need for additional task-specific heads. With minimal fine-tuning on all downstream tasks, our model significantly outperforms existing methods on benchmarks including ScanRefer, Multi3DRefer, Scan2Cap, ScanQA, and SQA3D.

We propose a novel method, \textbf{TwinAct}, to tackle the challenge of decoupling actions and actors in order to customize the text-guided diffusion models (TGDMs) for few-shot action image generation. TwinAct addresses the limitations of existing methods that struggle to decouple actions from other semantics (e.g., the actor's appearance) due to the lack of an effective inductive bias with few exemplar images. Our approach introduces a common action space, which is a textual embedding space focused solely on actions, enabling precise customization without actor-related details. Specifically, TwinAct involves three key steps: 1) Building common action space based on a set of representative action phrases; 2) Imitating the customized action within the action space; and 3) Generating highly adaptable customized action images in diverse contexts with action similarity loss. To comprehensively evaluate TwinAct, we construct a novel benchmark, which provides sample images with various forms of actions. Extensive experiments demonstrate TwinAct's superiority in generating accurate, context-independent customized actions while maintaining the identity consistency of different subjects, including animals, humans, and even customized actors.

Understanding human emotions is fundamental to enhancing human-computer interaction, especially for embodied agents that mimic human behavior. Traditional emotion analysis often takes a third-person perspective, limiting the ability of agents to interact naturally and empathetically. To address this gap, this paper presents $E^3$ for Exploring Embodied Emotion, the first massive first-person view video dataset. $E^3$ contains more than $50$ hours of video, capturing $8$ different emotion types in diverse scenarios and languages. The dataset features videos recorded by individuals in their daily lives, capturing a wide range of real-world emotions conveyed through visual, acoustic, and textual modalities. By leveraging this dataset, we define $4$ core benchmark tasks - emotion recognition, emotion classification, emotion localization, and emotion reasoning - supported by more than $80$k manually crafted annotations, providing a comprehensive resource for training and evaluating emotion analysis models. We further present Emotion-LlaMa, which complements visual modality with acoustic modality to enhance the understanding of emotion in first-person videos. The results of comparison experiments with a large number of baselines demonstrate the superiority of Emotion-LlaMa and set a new benchmark for embodied emotion analysis. We expect that $E^3$ can promote advances in multimodal understanding, robotics, and augmented reality, and provide a solid foundation for the development of more empathetic and context-aware embodied agents.

Video-to-audio (V2A) generation aims to synthesize content-matching audio from silent video, and it remains challenging to build V2A models with high generation quality, efficiency, and visual-audio temporal synchrony. We propose Frieren, a V2A model based on rectified flow matching. Frieren regresses the conditional transport vector field from noise to spectrogram latent with straight paths and conducts sampling by solving ODE, outperforming autoregressive and score-based models in terms of audio quality. By employing a non-autoregressive vector field estimator based on a feed-forward transformer and channel-level cross-modal feature fusion with strong temporal alignment, our model generates audio that is highly synchronized with the input video. Furthermore, through reflow and one-step distillation with guided vector field, our model can generate decent audio in a few, or even only one sampling step. Experiments indicate that Frieren achieves state-of-the-art performance in both generation quality and temporal alignment on VGGSound, with alignment accuracy reaching 97.22\%, and 6.2\% improvement in inception score over the strong diffusion-based baseline. Audio samples and code are available at http://frieren-v2a.github.io.

Talking face generation (TFG) aims to animate a target identity's face to create realistic talking videos. Personalized TFG is a variant that emphasizes the perceptual identity similarity of the synthesized result (from the perspective of appearance and talking style). While previous works typically solve this problem by learning an individual neural radiance field (NeRF) for each identity to implicitly store its static and dynamic information, we find it inefficient and non-generalized due to the per-identity-per-training framework and the limited training data. To this end, we propose MimicTalk, the first attempt that exploits the rich knowledge from a NeRF-based person-agnostic generic model for improving the efficiency and robustness of personalized TFG. To be specific, (1) we first come up with a person-agnostic 3D TFG model as the base model and propose to adapt it into a specific identity; (2) we propose a static-dynamic-hybrid adaptation pipeline to help the model learn the personalized static appearance and facial dynamic features; (3) To generate the facial motion of the personalized talking style, we propose an in-context stylized audio-to-motion model that mimics the implicit talking style provided in the reference video without information loss by an explicit style representation. The adaptation process to an unseen identity can be performed in 15 minutes, which is 47 times faster than previous person-dependent methods. Experiments show that our MimicTalk surpasses previous baselines regarding video quality, efficiency, and expressiveness. Video samples are available at https://mimictalk.github.io .

Motion-to-music and music-to-motion have been studied separately, each attracting substantial research interest within their respective domains. The interaction between human motion and music is a reflection of advanced human intelligence, and establishing a unified relationship between them is particularly important. However, to date, there has been no work that considers them jointly to explore the modality alignment within. To bridge this gap, we propose a novel framework, termed MoMu-Diffusion, for long-term and synchronous motion-music generation. Firstly, to mitigate the huge computational costs raised by long sequences, we propose a novel Bidirectional Contrastive Rhythmic Variational Auto-Encoder (BiCoR-VAE) that extracts the modality-aligned latent representations for both motion and music inputs. Subsequently, leveraging the aligned latent spaces, we introduce a multi-modal diffusion Transformer model and a cross-guidance sampling strategy to enable various generation tasks, including cross-modal, multi-modal, and variable-length generation. Extensive experiments demonstrate that MoMu-Diffusion surpasses recent state-of-the-art methods both qualitatively and quantitatively, and can synthesize realistic, diverse, long-term, and beat-matched music or motion sequences. The generated motion-music samples are available at https://momu-diffusion.github.io/.

Multi-modal contrastive representation (MCR) of more than three modalities is critical in multi-modal learning. Although recent methods showcase impressive achievements, the high dependence on large-scale, high-quality paired data and the expensive training costs limit their further development. Inspired by recent C-MCR, this paper proposes $\textbf{Ex}$tending $\textbf{M}$ultimodal $\textbf{C}$ontrastive $\textbf{R}$epresentation (Ex-MCR), a training-efficient and paired-data-free method to build unified contrastive representation for many modalities. Since C-MCR is designed to learn a new latent space for the two non-overlapping modalities and projects them onto this space, a significant amount of information from their original spaces is lost in the projection process. To address this issue, Ex-MCR proposes to extend one modality's space into the other's, rather than mapping both modalities onto a completely new space. This method effectively preserves semantic alignment in the original space. Experimentally, we extend pre-trained audio-text and 3D-image representations to the existing vision-text space. Without using paired data, Ex-MCR achieves comparable performance to advanced methods on a series of audio-image-text and 3D-image-text tasks and achieves superior performance when used in parallel with data-driven methods. Moreover, semantic alignment also emerges between the extended modalities (e.g., audio and 3D).

The scarcity of high-quality and multi-task singing datasets significantly hinders the development of diverse controllable and personalized singing tasks, as existing singing datasets suffer from low quality, limited diversity of languages and singers, absence of multi-technique information and realistic music scores, and poor task suitability.To tackle these problems, we present GTSinger, a large Global, multi-Technique, free-to-use, high-quality singing corpus with realistic music scores, designed for all singing tasks, along with its benchmarks.Particularly,(1) we collect 80.59 hours of high-quality singing voices, forming the largest recorded singing dataset;(2) 20 professional singers across nine widely spoken languages offer diverse timbres and styles;(3) we provide controlled comparison and phoneme-level annotations of six commonly used singing techniques, helping technique modeling and control;(4) GTSinger offers realistic music scores, assisting real-world musical composition;(5) singing voices are accompanied by manual phoneme-to-audio alignments, global style labels, and 16.16 hours of paired speech for various singing tasks.Moreover, to facilitate the use of GTSinger, we conduct four benchmark experiments: technique-controllable singing voice synthesis, technique recognition, style transfer, and speech-to-singing conversion.

Recent methods for video question answering (VideoQA), aiming to generate answers based on given questions and video content, have made significant progress in cross-modal interaction. From the perspective of video understating, these existing frameworks concentrate on the various levels of visual content, partially assisted by subtitles. However, audio information is also instrumental in helping get correct answers, especially in videos with real-life scenarios. Indeed, in some cases, both audio and visual contents are required and complement each other to answer questions, which is defined as audio-visual question answering (AVQA). In this paper, we focus on importing raw audio for AVQA and contribute in three ways. Firstly, due to no dataset annotating QA pairs for raw audio, we introduce E-AVQA, a manually annotated and large-scale dataset involving multiple modalities. E-AVQA consists of 34,033 QA pairs on 33,340 clips of 18,786 videos from the e-commerce scenarios. Secondly, we propose a multi-granularity relational attention method with contrastive constraints between audio and visual features after the interaction, named MGN, which captures local sequential representation by leveraging the pairwise potential attention mechanism and obtains global multi-modal representation via designing the novel ternary potential attention mechanism. Thirdly, our proposed MGN outperforms the baseline on dataset E-AVQA, achieving 20.73% on [email protected] and 19.81% on BLEU@1, demonstrating its superiority with at least 1.02 improvement on [email protected] and about 10% on timing complexity over the baseline.

Recommendation performance usually exhibits a long-tail distribution over users — a small portion of head users enjoy much more accurate recommendation services than the others. We reveal two sources of this performance heterogeneity problem: the uneven distribution of historical interactions (a natural source); and the biased training of recommender models (a model source). As addressing this problem cannot sacrifice the overall performance, a wise choice is to eliminate the model bias while maintaining the natural heterogeneity. The key to debiased training lies in eliminating the effect of confounders that influence both the user's historical behaviors and the next behavior. The emerging causal recommendation methods achieve this by modeling the causal effect between user behaviors, however potentially neglect unobserved confounders (e.g., friend suggestions) that are hard to measure in practice. To address unobserved confounders, we resort to the front-door adjustment (FDA) in causal theory and propose a causal multi-teacher distillation framework (CausalD). FDA requires proper mediators in order to estimate the causal effects of historical behaviors on the next behavior. To achieve this, we equip CausalD with multiple heterogeneous recommendation models to model the mediator distribution. Then, the causal effect estimated by FDA is the expectation of recommendation prediction over the mediator distribution and the prior distribution of historical behaviors, which is technically achieved by multi-teacher ensemble. To pursue efficient inference, CausalD further distills multiple teachers into one student model to directly infer the causal effect for making recommendations. We instantiate CausalD on two representative models, DeepFM and DIN, and conduct extensive experiments on three real-world datasets, which validate the superiority of CausalD over state-of-the-art methods. Through in-depth analysis, we find that CausalD largely improves the performance of tail users, reduces the performance heterogeneity, and enhances the overall performance.

Video moment retrieval (VMR) aims to localize the target moment in an untrimmed video according to the given nature language query. The existing algorithms typically rely on clean annotations to train their models. However, making annotations by human labors may introduce much noise. Thus, the video moment retrieval models will not be well trained in practice. In this article, we present a simple yet effective video moment retrieval framework via bottom-up schema, which is in end-to-end manners and robust to noisy label training. Specifically, we extract the multimodal features by syntactic graph convolutional networks and multihead attention layers, which are fused by the cross gates and the bilinear approach. Then, the feature pyramid networks are constructed to encode plentiful scene relationships and capture high semantics. Furthermore, to mitigate the effects of noisy annotations, we devise the multilevel losses characterized by two levels: a frame-level loss that improves noise tolerance and an instance-level loss that reduces adverse effects of negative instances. For the frame level, we adopt the Gaussian smoothing to regard noisy labels as soft labels through the partial fitting. For the instance level, we exploit a pair of structurally identical models to let them teach each other during iterations. This leads to our proposed robust video moment retrieval model, which experimentally and significantly outperforms the state-of-the-art approaches on standard public datasets ActivityCaption and textually annotated cooking scene (TACoS). We also evaluate the proposed approach on the different manual annotation noises to further demonstrate the effectiveness of our model.

In recommender systems, click behaviors play a fundamental role in mining users’ interests and training models (clicked items as positive samples). Such signals are implicit feedback and are arguably less representative of users’ inherent interests. Most existing works denoise implicit feedback by introducing external signals, such as gaze, dwell time, and “like” behaviors. However, such explicit feedback is not always routinely available, or might be problematic to collect on a large scale. In this paper, we identify that an interaction’s related structural patterns in its neighborhood graph are potentially correlated with some outcome of implicit feedback (i.e., users’ ratings after consuming items), analogous to findings in other domains such as social networks. Inspired by this finding, we propose a novel Structure LEarning based Denoising (SLED) framework for denoising recommendation without explicit signals, which consists of two phases: center-aware graph structure learning and denoised recommendation . Phase 1 pre-trains a structural encoder in a self-supervised manner and learns to capture an interaction’s related structural patterns in its neighborhood graph. Phase 2 transfers the structure encoder to downstream recommendation datasets, which helps to down-weight the effect of noisy interactions on user interest modeling and loss calculation. We collect a relatively noisy industrial dataset across several days during a period of product promotion festival. Extensive experiments on this dataset and multiple public datasets demonstrate that the proposed SLED framework can significantly improve the recommendation quality over various base recommendation models.

Existing video-audio understanding models are trained and evaluated in an intra-domain setting, facing performance degeneration in real-world applications where multiple domains and distribution shifts naturally exist. The key to video-audio domain generalization (VADG) lies in alleviating spurious correlations over multi-modal features. To achieve this goal, we resort to causal theory and attribute such correlation to confounders affecting both video-audio features and labels. We propose a DeVADG framework that conducts uni-modal and cross-modal deconfounding through back-door adjustment. DeVADG performs cross-modal disentanglement and obtains fine-grained confounders at both class-level and domain-level using half-sibling regression and unpaired domain transformation, which essentially identifies domain-variant factors and class-shared factors that cause spurious correlations between features and false labels. To promote VADG research, we collect a VADG-Action dataset for video-audio action recognition with over 5,000 video clips across four domains (e.g., cartoon and game) and ten action classes (e.g., cooking and riding). We conduct extensive experiments, i.e., multi-source DG, single-source DG, and qualitative analysis, validating the rationality of our causal analysis and the effectiveness of the DeVADG framework.

Diffusion models have recently exhibited remarkable abilities to synthesize striking image samples since the introduction of denoising diffusion probabilistic models (DDPMs). Their key idea is to disrupt images into noise through a fixed forward process and learn its reverse process to generate samples from noise in a denoising way. For conditional DDPMs, most existing practices relate conditions only to the reverse process and fit it to the reversal of unconditional forward process. We find this will limit the condition modeling and generation in a small time window. In this paper, we propose a novel and flexible conditional diffusion model by introducing conditions into the forward process. We utilize extra latent space to allocate an exclusive diffusion trajectory for each condition based on some shifting rules, which will disperse condition modeling to all timesteps and improve the learning capacity of model. We formulate our method, which we call ShiftDDPMs, and provide a unified point of view on existing related methods. Extensive qualitative and quantitative experiments on image synthesis demonstrate the feasibility and effectiveness of ShiftDDPMs.

Speech Recognition builds a bridge between the multimedia streaming (audio-only, visualonly or audio-visual) and the corresponding text transcription.However, when training the specific model of new domain, it often gets stuck in the lack of new-domain utterances, especially the labeled visual utterances.To break through this restriction, we attempt to achieve zero-shot modality transfer by maintaining the multi-modality alignment in phoneme space learned with unlabeled multimedia utterances in the high resource domain during the pretraining (Shi et al., 2022), and propose a training system Open-modality Speech Recognition (OpenSR) that enables the models trained on a single modality (e.g., audio-only) applicable to more modalities (e.g., visual-only and audio-visual).Furthermore, we employ a cluster-based prompt tuning strategy to handle the domain shift for the scenarios with only common words in the new domain utterances.We demonstrate that OpenSR enables modality transfer from one to any in three different settings (zero-, few-and fullshot), and achieves highly competitive zeroshot performance compared to the existing fewshot and full-shot lip-reading methods.To the best of our knowledge, OpenSR achieves the state-of-the-art performance of word error rate in LRS2 on audio-visual speech recognition and lip-reading with 2.7% and 25.0%, respectively.The code and demo are available at https://github.com/Exgc/OpenSR.

Rongjie Huang, Huadai Liu, Xize Cheng, Yi Ren, Linjun Li, Zhenhui Ye, Jinzheng He, Lichao Zhang, Jinglin Liu, Xiang Yin, Zhou Zhao. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Mengze Li, Tianbao Wang, Jiahe Xu, Kairong Han, Shengyu Zhang, Zhou Zhao, Jiaxu Miao, Wenqiao Zhang, Shiliang Pu, Fei Wu. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Audio-visual text generation aims to understand multi-modality contents and translate them into texts.Although various transfer learning techniques of text generation have been proposed, they focused on uni-modal analysis (e.g., text-to-text, visual-to-text) and lack consideration of multi-modal content and cross-modal relation.Motivated by the fact that humans can recognize the timbre of the same low-level concepts (e.g., footstep, rainfall, and laughing), even in different visual conditions, we aim to mitigate the domain discrepancies by audiovisual correlation.In this paper, we propose a novel Transferable Audio-Visual Text Generation framework, named TAVT, which consists of two key components: Audio-Visual Meta-Mapper (AVMM) and Dual Counterfactual Contrastive Learning (DCCL).(1) AVMM first introduces a universal auditory semantic space and drifts the domain-invariant low-level concepts into visual prefixes.Then the reconstructbased learning encourages the AVMM to learn "which pixels belong to the same sound" and achieve audio-enhanced visual prefix.The welltrained AVMM can be further applied to unimodal setting.(2) Furthermore, DCCL leverages the destructive counterfactual transformations to provide cross-modal constraints for AVMM from the perspective of feature distribution and text generation.(3) The experimental results show that TAVT outperforms the stateof-the-art methods across multiple domains (cross-datasets, cross-categories) and various modal settings (uni-modal, multi-modal).

The task of spoken video grounding aims to localize moments in videos that are relevant to descriptive spoken queries.However, extracting semantic information from speech and modeling the cross-modal correlation pose two critical challenges.Previous studies solve them by representing spoken queries based on the matched video frames, which require tremendous effort for frame-level labeling.In this work, we investigate weakly-supervised spoken video grounding, i.e., learning to localize moments without expensive temporal annotations.To effectively represent the cross-modal semantics, we propose Semantic Interaction Learning (SIL), a novel framework consisting of the acoustic-semantic pre-training (ASP) and acoustic-visual contrastive learning (AVCL).In ASP, we pre-train an effective encoder for the grounding task with three comprehensive tasks, where the robustness task enhances stability by explicitly capturing the invariance between time-and frequency-domain features, the conciseness task avoids over-smooth attention by compressing long sequence into segments, and the semantic task improves spoken language understanding by modeling the precise semantics.In AVCL, we mine pseudo labels with discriminative sampling strategies and directly strengthen the interaction between speech and video by maximizing their mutual information.Extensive experiments demonstrate the effectiveness and superiority of our method.1

Zhenhui Ye, Rongjie Huang, Yi Ren, Ziyue Jiang, Jinglin Liu, Jinzheng He, Xiang Yin, Zhou Zhao. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Building benchmarks to systemically analyze different capabilities of video question answering (VideoQA) models is challenging yet crucial. Existing benchmarks often use non-compositional simple questions and suffer from language biases, making it difficult to diagnose model weaknesses incisively. A recent benchmark AGQA [8] poses a promising paradigm to generate QA pairs automatically from pre-annotated scene graphs, enabling it to measure diverse reasoning abilities with granular control. However, its questions have limitations in reasoning about the fine-grained semantics in videos as such information is absent in its scene graphs. To this end, we present ANetQA, a large-scale benchmark that supports fine-grained compositional reasoning over the challenging untrimmed videos from ActivityNet [4]. Similar to AGQA, the QA pairs in ANetQA are automatically generated from annotated video scene graphs. The fine-grained properties of ANetQA are reflected in the following: (i) untrimmed videos with fine-grained semantics; (ii) spatio-temporal scene graphs with fine-grained taxonomies; and (iii) diverse questions generated from fine-grained templates. ANetQA attains 1.4 billion unbalanced and 13.4 million balanced QA pairs, which is an order of magnitude larger than AGQA with a similar number of videos. Comprehensive experiments are performed for state-of-the-art methods. The best model achieves 44.5% accuracy while human performance tops out at 84.5%, leaving sufficient room for improvement.

Product Retrieval (PR) and Grounding (PG), aiming to seek image and object-level products respectively according to a textual query, have attracted great interest recently for better shopping experience. Owing to the lack of relevant datasets, we collect two large-scale benchmark datasets from Taobao Mall and Live domains with about 474k and 101k image-query pairs for PR, and manually annotate the object bounding boxes in each image for PG. As annotating boxes is expensive and time-consuming, we attempt to transfer knowledge from annotated domain to unannotated for PG to achieve unsupervised Domain Adaptation (PG-DA). We propose a Domain Adaptive Product Seeker (DATE) framework, regarding PR and PG as Product Seeking problem at different levels, to assist the query date the product. Concretely, we first design a semantics-aggregated feature extractor for each modality to obtain concentrated and comprehensive features for following efficient retrieval and fine-grained grounding tasks. Then, we present two cooperative seekers to simultaneously search the image for PR and localize the product for PG. Besides, we devise a domain aligner for PG-DA to alleviate unimodal marginal and multi-modal conditional distribution shift between source and target domains, and design a pseudo box generator to dynamically select reliable instances and generate bounding boxes for further knowledge transfer. Extensive experiments show that our DATE achieves satisfactory performance in fully-supervised PR, PG and unsupervised PG-DA. Our desensitized datasets will be publicly available here11https://github.com/Taobao-live/Product-Seeking.

Spatio-temporal video grounding aims to localize the aligned visual tube corresponding to a language query. Existing techniques achieve such alignment by exploiting dense boundary and bounding box annotations, which can be prohibitively expensive. To bridge the gap, we investigate the weakly-supervised setting, where models learn from easily accessible video-language data without annotations. We identify that intra-sample spurious correlations among video-language components can be alleviated if the model captures the decomposed structures of video and language data. In this light, we propose a novel framework, namely WINNER, for hierarchical video-text understanding. WINNER first builds the language decomposition tree in a bottom-up manner, upon which the structural attention mechanism and top-down feature backtracking jointly build a multi-modal decomposition tree, permitting a hierarchical understanding of unstructured videos. The multi-modal decomposition tree serves as the basis for multi-hierarchy language-tube matching. A hierarchical contrastive learning objective is proposed to learn the multi-hierarchy correspondence and distinguishment with intra-sample and inter-sample video-text decomposition structures, achieving video-language decomposition structure alignment. Extensive experiments demonstrate the rationality of our design and its effectiveness beyond state-of-the-art weakly supervised methods, even some supervised methods.

Most sign language translation (SLT) methods to date require the use of gloss annotations to provide additional supervision information, however, the acquisition of gloss is not easy. To solve this problem, we first perform an analysis of existing models to confirm how gloss annotations make SLT easier. We find that it can provide two aspects of information for the model, 1) it can help the model implicitly learn the location of semantic boundaries in continuous sign language videos, 2) it can help the model understand the sign language video globally. We then propose gloss attention, which enables the model to keep its attention within video segments that have the same semantics locally, just as gloss helps existing models do. Furthermore, we transfer the knowledge of sentence-to-sentence similarity from the natural language model to our gloss attention SLT network (GASLT) to help it understand sign language videos at the sentence level. Experimental results on multiple large-scale sign language datasets show that our proposed GASLT model significantly outperforms existing methods. Our code is provided in https://github.com/YinAoXiong/GASLT.

Mengze Li, Tianqi Zhao, Bai Jionghao, Baoyi He, Jiaxu Miao, Wei Ji, Zheqi Lv, Zhou Zhao, Shengyu Zhang, Wenqiao Zhang, Fei Wu. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Text-to-speech(TTS) has undergone remarkable improvements in performance, particularly with the advent of Denoising Diffusion Probabilistic Models (DDPMs).However, the perceived quality of audio depends not solely on its content, pitch, rhythm, and energy, but also on the physical environment.In this work, we propose ViT-TTS, the first visual TTS model with scalable diffusion transformers.ViT-TTS complement the phoneme sequence with the visual information to generate high-perceived audio, opening up new avenues for practical applications of AR and VR to allow a more immersive and realistic audio experience.To mitigate the data scarcity in learning visual acoustic information, we 1) introduce a selfsupervised learning framework to enhance both the visual-text encoder and denoiser decoder; 2) leverage the diffusion transformer scalable in terms of parameters and capacity to learn visual scene information.Experimental results demonstrate that ViT-TTS achieves new stateof-the-art results, outperforming cascaded systems and other baselines regardless of the visibility of the scene.With low-resource data (1h, 2h, 5h), ViT-TTS achieves comparative results with rich-resource baselines.1 2

3D visual grounding aims to localize the target object in a 3D point cloud by a free-form language description.Typically, the sentences describing the target object tend to provide information about its relative relation between other objects and its position within the whole scene.In this work, we propose a relation-aware onestage framework, named 3D Relative Positionaware Network (3DRP-Net), which can effectively capture the relative spatial relationships between objects and enhance object attributes.Specifically, 1) we propose a 3D Relative Position Multi-head Attention (3DRP-MA) module to analyze relative relations from different directions in the context of object pairs, which helps the model to focus on the specific object relations mentioned in the sentence.2) We designed a soft-labeling strategy to alleviate the spatial ambiguity caused by redundant points, which further stabilizes and enhances the learning process through a constant and discriminative distribution.Extensive experiments conducted on three benchmarks (i.e., ScanRefer and Nr3D/Sr3D) demonstrate that our method outperforms all the state-of-the-art methods in general.

Video-to-sound generation aims to generate realistic and natural sound given a video input. However, previous video-to-sound generation methods can only generate a random or average timbre without any controls of the generated sound timbre, leading to the problem that people cannot obtain the desired timbre under these methods sometimes. In this paper, we propose the task of generating sound with a specific timbre given a silent video input and a reference audio sample. To solve this task, we first use three encoders to disentangle each target sound audio into temporal, acoustic, and background information respectively, then we use a decoder to reconstruct the audio given these disentangled representations. To make the generated result achieve better quality and temporal alignment, we also adopt a mel discriminator and a temporal discriminator for the adversarial training. Our experimental results on the VAS dataset demonstrate that our method can generate high-quality audio samples with good synchronization with events in video and high timbre similarity with the reference audio. Our demos have been published on https://conferencedemos.github.io/icassp23/.

ICASSP2023 General Meeting Understanding and Generation Challenge (MUG) focuses on prompting a wide range of spoken language processing (SLP) research on meeting transcripts, as SLP applications are critical to improve users’ efficiency in grasping important information in meetings. MUG includes five tracks, including topic segmentation, topic-level and session-level extractive summarization, topic title generation, keyphrase extraction, and action item detection. To facilitate MUG, we construct and release a large-scale meeting dataset, the AliMeeting4MUG Corpus. We review the dataset, track settings and baselines, and summarize the challenge results and major techniques used in the submissions.

Listening to long video/audio recordings from video conferencing and online courses for acquiring information is extremely inefficient. Even after ASR systems transcribe recordings into long-form spoken language documents, reading ASR transcripts only partly speeds up seeking information. It has been observed that a range of NLP applications, such as keyphrase extraction, topic segmentation, and summarization, significantly improve users’ efficiency in grasping important information. The meeting scenario is among the most valuable scenarios for deploying these spoken language processing (SLP) capabilities. However, the lack of large-scale public meeting datasets annotated for these SLP tasks severely hinders their advancement. To prompt SLP advancement, we establish a large-scale general Meeting Understanding and Generation Benchmark (MUG) to benchmark the performance of a wide range of SLP tasks, including topic segmentation, topic-level and session-level extractive summarization and topic title generation, keyphrase extraction, and action item detection. To facilitate the MUG benchmark, we construct and release a large-scale meeting dataset for comprehensive long-form SLP development, the AliMeeting4MUG Corpus, which consists of 654 recorded Mandarin meeting sessions with diverse topic coverage, with manual annotations for SLP tasks on manual transcripts of meeting recordings. To the best of our knowledge, the AliMeeting4MUG Corpus is so far the largest meeting corpus in scale and facilitates most SLP tasks. In this paper, we provide a detailed introduction of this corpus, SLP tasks and evaluation methods, baseline systems and their performance1.

Multi-media communications facilitate global interaction among people. However, despite researchers exploring cross-lingual translation techniques such as machine translation and audio speech translation to overcome language barriers, there is still a shortage of cross-lingual studies on visual speech. This lack of research is mainly due to the absence of datasets containing visual speech and translated text pairs. In this paper, we present AVMuST-TED, the first dataset for Audio-Visual Multilingual Speech Translation, derived from TED talks. Nonetheless, visual speech is not as distinguishable as audio speech, making it difficult to develop a mapping from source speech phonemes to the target language text. To address this issue, we propose MixSpeech, a cross-modality self-learning framework that utilizes audio speech to regularize the training of visual speech tasks. To further minimize the cross-modality gap and its impact on knowledge transfer, we suggest adopting mixed speech, which is created by interpolating audio and visual streams, along with a curriculum learning strategy to adjust the mixing ratio as needed. MixSpeech enhances speech translation in noisy environments, improving BLEU scores for four languages on AVMuST-TED by +1.4 to +4.2. Moreover, it achieves state-of-the-art performance in lip reading on CMLR (11.1%), LRS2 (25.5%), and LRS3 (28.0%).

Current video captioning efforts most focus on describing a single video while the need for captioning videos in groups has increased considerably. In this study, we propose a new task, group video captioning, which aims to infer the desired content among a group of target videos and describe it with another group of related reference videos. This task requires the model to effectively summarize the target videos and accurately describe the distinguishing content compared to the reference videos, and it becomes more difficult as the video length increases. To solve this problem, 1) First, we propose an efficient relational approximation (ERA) to identify the shared content among videos while the complexity is linearly related to the number of videos. 2) Then, we introduce a contextual feature refinery with intra-group self-supervision to capture the contextual information and further refine the common properties. 3) In addition, we construct two group video captioning datasets derived from the YouCook2 and the ActivityNet Captions. The experimental results demonstrate the effectiveness of our method on this new task.

3D visual grounding involves finding a target object in a 3D scene that corresponds to a given sentence query. Although many approaches have been proposed and achieved impressive performance, they all require dense object-sentence pair annotations in 3D point clouds, which are both time-consuming and expensive. To address the problem that fine-grained annotated data is difficult to obtain, we propose to leverage weakly supervised annotations to learn the 3D visual grounding model, i.e., only coarse scene-sentence correspondences are used to learn object-sentence links. To accomplish this, we design a novel semantic matching model that analyzes the semantic similarity between object proposals and sentences in a coarse-to-fine manner. Specifically, we first extract object proposals and coarsely select the top-K candidates based on feature and class similarity matrices. Next, we reconstruct the masked keywords of the sentence using each candidate one by one, and the reconstructed accuracy finely reflects the semantic similarity of each candidate to the query. Additionally, we distill the coarse-to-fine semantic matching knowledge into a typical two-stage 3D visual grounding model, which reduces inference costs and improves performance by taking full advantage of the well-studied structure of the existing architectures. We conduct extensive experiments on ScanRefer, Nr3D, and Sr3D, which demonstrate the effectiveness of our proposed method.

Existing open vocabulary object detection (OVD) works expand the object detector toward open categories by replacing the classifier with the category text embeddings and optimizing the region-text alignment on data of the base categories. However, both the class-agnostic proposal generator and the classifier are biased to the seen classes as demonstrated by the gaps of objectness and accuracy assessment between base and novel classes. In this paper, an open corpus, composed of a set of external object concepts and clustered to several centroids, is introduced to improve the generalization ability in the detector. We propose the generalized objectness assessment (GOAT) in the proposal generator based on the visual-text alignment, where the similarities of visual feature to the cluster centroids are summarized as the objectness. This simple heuristic evaluates objectness with concepts in open corpus and is thus generalized to open categories. We further propose category expanding (CE) with open corpus in two training tasks, which enables the detector to perceive more categories in the feature space and get more reasonable optimization direction. For the classification task, we introduce an open corpus classifier by reconstructing original classifier with similar words in text space. For the image-caption alignment task, the open corpus centroids are incorporated to enlarge the negative samples in the contrastive loss. Extensive experiments demonstrate the effectiveness of GOAT and CE, which greatly improve the performance on novel classes and get new state-of-the-art on the OVD benchmarks.

Large-scale multimodal generative modeling has created milestones in text-to-image and text-to-video generation. Its application to audio still lags behind for two main reasons: the lack of large-scale datasets with high-quality text-audio pairs, and the complexity of modeling long continuous audio data. In this work, we propose Make-An-Audio with a prompt-enhanced diffusion model that addresses these gaps by 1) introducing pseudo prompt enhancement with a distill-then-reprogram approach, it alleviates data scarcity with orders of magnitude concept compositions by using language-free audios; 2) leveraging spectrogram autoencoder to predict the self-supervised audio representation instead of waveforms. Together with robust contrastive language-audio pretraining (CLAP) representations, Make-An-Audio achieves state-of-the-art results in both objective and subjective benchmark evaluation. Moreover, we present its controllability and generalization for X-to-Audio with "No Modality Left Behind", for the first time unlocking the ability to generate high-definition, high-fidelity audios given a user-defined modality input. Audio samples are available at https://Make-An-Audio.github.io

Unsupervised text style transfer task aims to rewrite a text into target style while preserving its main content. Traditional methods rely on the use of a fixed-sized vector to regulate text style, which is difficult to accurately convey the style strength for each individual token. In fact, each token of a text contains different style intensity and makes different contribution to the overall style. Our proposed method addresses this issue by assigning individual style vector to each token in a text, allowing for fine-grained control and manipulation of the style strength. Additionally, an adversarial training framework integrated with teacher-student learning is introduced to enhance training stability and reduce the complexity of high-dimensional optimization. The results of our experiments demonstrate the efficacy of our method in terms of clearly improved style transfer accuracy and content preservation in both two-style transfer and multi-style transfer settings.

Though previous works have shown remarkable achievements in singing voice generation, most existing models focus on one specific application and there is a lack of unified singing voice synthesis models. In addition to low relevance among tasks, different input modalities are one of the most intractable hindrances. Current methods suffer from information confusion and they can not perform precise control. In this work, we propose UniSinger, a unified end-to-end singing voice synthesizer, which integrates three abilities related to singing voice generation: singing voice synthesis (SVS), singing voice conversion (SVC), and singing voice editing (SVE) into a single framework. Specifically, we perform representation disentanglement for controlling different attributes of the singing voice. We further propose a cross-modality information matching method to close the distribution gap between multi-modal inputs and achieve end-to-end training. The experiments conducted on the OpenSinger dataset demonstrate that UniSinger achieves state-of-the-art results in three applications. Further extensive experiments verify the capability of representation disentanglement and information matching, reflecting that UniSinger enjoys great superiority in sample quality, timbre similarity, and multi-task compatibility. Audio samples can be found in https://unisinger.github.io/Samples/.

Conventional pipeline of multimodal learning consists of three stages, including encoding, fusion, and decoding. Most existing methods under missing modality condition focus on the first stage and aim to learn the modality invariant representation or reconstruct missing features. However, these methods rely on strong assumptions (i.e., all the pre-defined modalities are available for each input sample during training and the number of modalities is fixed). To solve this problem, we propose a simple yet effective method called Interaction Augmented Prototype Decomposition (IPD) for a more general setting, where the number of modalities is arbitrary and there are various incomplete modality conditions happening in both training and inference phases, even there are unseen testing conditions. Different from the previous methods, we improve the decoding stage. Concretely, IPD jointly learns the common and modality-specific task prototypes. Considering that the number of missing modality conditions scales exponentially with the number of modalities O(2n) and different conditions may have implicit interaction, the low-rank partial prototype decomposition with enough theoretical analysis is employed for modality-specific components to reduce the complexity. The decomposition also can promote unseen generalization with the modality factors of existing conditions. To simulate the low-rank setup, we further constrain the explicit interaction of specific modality conditions by employing disentangled contrastive constraints. Extensive results on the newly-created benchmarks of multiple tasks illustrate the effectiveness of our proposed model.

This paper addresses the Unsupervised Domain Adaptation (UDA) for the dense frame prediction task - Video Object Grounding (VOG). This investigation springs from the recognition of the limited generalization capabilities of data-driven approaches when confronted with unseen test scenarios. We set the goal of enhancing the adaptability of the source-dominated model from a labeled domain to the unlabeled target domain through re-training on pseudo-labels (i.e., predicted boxes of language-described objects). Given the potential for source-domain biases in the pseudo-label generation, we decompose the labeling refinement as two cascaded debiasing subroutines: (1) we develop a discarded training strategy to correct the Biased Proposal Selection by filtering out the examples with uncertain proposals selected from the proposal (candidate box) set. The identifier of these uncertain examples is the discordance between the predictions of the source-dominated model and those of a target-domain clustered classifier, which remains free from the source-domain bias. (2) With the refined proposals as a foundation, we measure Grounding Coordinate Offset based on the semantic distance of the model's prediction across domains, based on which we alleviate source-domain bias in the target model through adversarial learning. To verify the superiority of the proposed method, we collected two UDA-VOG datasets called I2O-VOG and R2M-VOG by manually dividing and combining the well-known VOG datasets. The extensive experiments on them show our model significantly outperforms SOTA methods by a large margin.

In this paper, we study the task of referring semantic segmentation in a highly practical setting, in which labeled visual data with corresponding text descriptions are available in the source, but only unlabeled visual data (without text descriptions) are available in the target. It is a challenging task that has many difficulties: (1) how to obtain proper queries for the target domain; (2) how to adapt visual-text joint distribution shifts; (3) how to maintain the original segmentation performance. Thus, we propose a cycle-consistent vision-language matching network to narrow down the domain gap and ease adaptation difficulty. Our model has significant practical applications since they are capable generalising to new data sources without requiring corresponding text annotations. First, a pseudo-text selector is devised to handle the missing modality, through the pre-trained clip model to measure the gap between query features of the source and visual features of the target. Next, a cross-domain segmentation predictor is adopted, which prompts the joint representations to be domain invariant and minimize the discrepancy between two domains. Then, we present a cycle-consistent query matcher to learn discriminative features via reconstructing visual features from masks. Instead of doing the textual comparison, we match the visual features to the pseudo queries. Extensive experiments show the effectiveness of our method.

In recent years, the deployment of large-scale pre-trained models in audio-visual downstream tasks has yielded remarkable outcomes. However, these models, primarily trained on single-modality unconstrained datasets, still encounter challenges in feature extraction for multi-modal tasks, leading to suboptimal performance. This limitation arises due to the introduction of irrelevant modality-specific information during encoding, which adversely affects the performance of downstream tasks. To address this challenge, this paper proposes a novel Dual-Guided Spatial-Channel-Temporal (DG-SCT) attention mechanism. This mechanism leverages audio and visual modalities as soft prompts to dynamically adjust the parameters of pre-trained models based on the current multi-modal input features. Specifically, the DG-SCT module incorporates trainable cross-modal interaction layers into pre-trained audio-visual encoders, allowing adaptive extraction of crucial information from the current modality across spatial, channel, and temporal dimensions, while preserving the frozen parameters of large-scale pre-trained models. Experimental evaluations demonstrate that our proposed model achieves state-of-the-art results across multiple downstream tasks, including AVE, AVVP, AVS, and AVQA. Furthermore, our model exhibits promising performance in challenging few-shot and zero-shot scenarios. The source code and pre-trained models are available at https://github.com/haoyi-duan/DG-SCT.

The focus of our work is on diagnostic tasks in personalized learning, such as cognitive diagnosis and knowledge tracing. The goal of these tasks is to assess students' latent proficiency on knowledge concepts through analyzing their historical learning records. However, existing research has been limited to single-course scenarios; cross-course studies have not been explored due to a lack of dataset. We address this issue by constructing PTADisc, a Diverse, Immense, Student-centered dataset that emphasizes its sufficient Cross-course information for personalized learning. PTADisc includes 74 courses, 1,530,100 students, 4,054 concepts, 225,615 problems, and over 680 million student response logs. Based on PTADisc, we developed a model-agnostic Cross-Course Learner Modeling Framework (CCLMF) which utilizes relationships between students' proficiency across courses to alleviate the difficulty of diagnosing student knowledge state in cold-start scenarios. CCLMF uses a meta network to generate personalized mapping functions between courses. The experimental results on PTADisc verify the effectiveness of CCLMF with an average improvement of 4.2% on AUC. We also report the performance of baseline models for cognitive diagnosis and knowledge tracing over PTADisc, demonstrating that our dataset supports a wide scope of research in personalized learning. Additionally, PTADisc contains valuable programming logs and student-group information that are worth exploring in the future.

Multi-modal Contrastive Representation (MCR) learning aims to encode different modalities into a semantically aligned shared space. This paradigm shows remarkable generalization ability on numerous downstream tasks across various modalities. However, the reliance on massive high-quality data pairs limits its further development on more modalities. This paper proposes a novel training-efficient method for learning MCR without paired data called Connecting Multi-modal Contrastive Representations (C-MCR). Specifically, given two existing MCRs pre-trained on $(\mathcal{A}$, $\mathcal{B})$ and $(\mathcal{B}$, $\mathcal{C})$ modality pairs, we project them to a new space and use the data from the overlapping modality $\mathcal{B}$ to aligning the two MCRs in the new space. Meanwhile, since the modality pairs $(\mathcal{A}$, $\mathcal{B})$ and $(\mathcal{B}$, $\mathcal{C})$ are already aligned within each MCR, the connection learned by overlapping modality can also be transferred to non-overlapping modality pair $(\mathcal{A}$, $\mathcal{C})$. To unleash the potential of C-MCR, we further introduce a semantic-enhanced inter- and intra-MCR connection method. We first enhance the semantic consistency and completion of embeddings across different modalities for more robust alignment. Then we utilize the inter-MCR alignment to establish the connection, and employ the intra-MCR alignment to better maintain the connection for inputs from non-overlapping modalities. To demonstrate the effectiveness of C-MCR, we take the field of audio-visual and 3D-language learning as examples. Specifically, we connect CLIP and CLAP via texts to derive audio-visual representations, and integrate CLIP and ULIP via images for 3D-language representations. Remarkably, without using any paired data, C-MCR for audio-visual achieves state-of-the-art performance on audio-image retrieval, audio-visual source localization, and counterfactual audio-image recognition tasks. Furthermore, C-MCR for 3D-language also attains advanced zero-shot 3D point cloud classification accuracy on ModelNet40. Our project page is available at \url{https://c-mcr.github.io/C-MCR/}

This paper introduces a novel task called Cross Modal Generalization (CMG), which addresses the challenge of learning a unified discrete representation from paired multimodal data during pre-training. Then in downstream tasks, the model can achieve zero-shot generalization ability in other modalities when only one modal is labeled. Existing approaches in multimodal representation learning focus more on coarse-grained alignment or rely on the assumption that information from different modalities is completely aligned, which is impractical in real-world scenarios. To overcome this limitation, we propose \textbf{Uni-Code}, which contains two key contributions: the Dual Cross-modal Information Disentangling (DCID) module and the Multi-Modal Exponential Moving Average (MM-EMA). These methods facilitate bidirectional supervision between modalities and align semantically equivalent information in a shared discrete latent space, enabling fine-grained unified representation of multimodal sequences. During pre-training, we investigate various modality combinations, including audio-visual, audio-text, and the tri-modal combination of audio-visual-text. Extensive experiments on various downstream tasks, i.e., cross-modal event classification, localization, cross-modal retrieval, query-based video segmentation, and cross-dataset event localization, demonstrate the effectiveness of our proposed methods. The code is available at https://github.com/haihuangcode/CMG.

Two-tower models are a prevalent matching framework for recommendation, which have been widely deployed in industrial applications. The success of two-tower matching attributes to its efficiency in retrieval among a large number of items, since the item tower can be precomputed and used for fast Approximate Nearest Neighbor (ANN) search. However, it suffers two main challenges, including limited feature interaction capability and reduced accuracy in online serving. Existing approaches attempt to design novel late interactions instead of dot products, but they still fail to support complex feature interactions or lose retrieval efficiency. To address these challenges, we propose a new matching paradigm named SparCode, which supports not only sophisticated feature interactions but also efficient retrieval. Specifically, SparCode introduces an all-to-all interaction module to model fine-grained query-item interactions. Besides, we design a discrete code-based sparse inverted index jointly trained with the model to achieve effective and efficient model inference. Extensive experiments have been conducted on open benchmark datasets to demonstrate the superiority of our framework. The results show that SparCode significantly improves the accuracy of candidate item matching while retaining the same level of retrieval efficiency with two-tower models.

In the field of music information retrieval (MIR), cover song identification (CSI) is a challenging task that aims to identify cover versions of a query song from a massive collection. Existing works still suffer from high intra-song variances and inter-song correlations, due to the entangled nature of version-specific and version-invariant factors in their modeling. In this work, we set the goal of disentangling version-specific and version-invariant factors, which could make it easier for the model to learn invariant music representations for unseen query songs. We analyze the CSI task in a disentanglement view with the causal graph technique, and identify the intra-version and inter-version effects biasing the invariant learning. To block these effects, we propose the disentangled music representation learning framework (DisCover) for CSI. DisCover consists of two critical components: (1) Knowledge-guided Disentanglement Module (KDM) and (2) Gradient-based Adversarial Disentanglement Module (GADM), which block intra-version and inter-version biased effects, respectively. KDM minimizes the mutual information between the learned representations and version-variant factors that are identified with prior domain knowledge. GADM identifies version-variant factors by simulating the representation transitions between intra-song versions, and exploits adversarial distillation for effect blocking. Extensive comparisons with best-performing methods and in-depth analysis demonstrate the effectiveness of DisCover and the and necessity of disentanglement for CSI.

In recommender systems, users' behavior data are driven by the interactions of user-item latent factors. To improve recommendation effectiveness and robustness, recent advances focus on latent factor disentanglement via variational inference. Despite significant progress, uncovering the underlying interactions, i.e., dependencies of latent factors, remains largely neglected by the literature. To bridge the gap, we investigate the joint disentanglement of user-item latent factors and the dependencies between them, namely latent structure learning. We propose to analyze the problem from the causal perspective, where a latent structure should ideally reproduce observational interaction data, and satisfy the structure acyclicity and dependency constraints, i.e., causal prerequisites. We further identify the recommendation-specific challenges for latent structure learning, i.e., the subjective nature of users' minds and the inaccessibility of private/sensitive user factors causing universally learned latent structure to be suboptimal for individuals. To address these challenges, we propose the personalized latent structure learning framework for recommendation, namely PlanRec, which incorporates 1) differentiable Reconstruction, Dependency, and Acyclicity regularizations to satisfy the causal prerequisites; 2) Personalized Structure Learning (PSL) which personalizes the universally learned dependencies through probabilistic modeling; and 3) uncertainty estimation which explicitly measures the uncertainty of structure personalization, and adaptively balances personalization and shared knowledge for different users. We conduct extensive experiments on two public benchmark datasets from MovieLens and Amazon, and a large-scale industrial dataset from Alipay. Empirical studies validate that PlanRec discovers effective shared/personalized structures, and successfully balances shared knowledge and personalization via rational uncertainty estimation.

Unconstrained lip-to-speech aims to generate corresponding speeches based on silent facial videos with no restriction to head pose or vocabulary. It is desirable to generate intelligible and natural speech with a fast speed in unconstrained settings. Currently, to handle the more complicated scenarios, most existing methods adopt the autoregressive architecture, which is optimized with the MSE loss. Although these methods have achieved promising performance, they are prone to bring issues including high inference latency and mel-spectrogram over-smoothness. To tackle these problems, we propose a novel flow-based non-autoregressive lip-to-speech model (GlowLTS) to break autoregressive constraints and achieve faster inference. Concretely, we adopt a flow-based decoder which is optimized by maximizing the likelihood of the training data and is capable of more natural and fast speech generation. Moreover, we devise a condition module to improve the intelligibility of generated speech. We demonstrate the superiority of our proposed method through objective and subjective evaluation on Lip2Wav-Chemistry-Lectures and Lip2Wav-Chess-Analysis datasets. Our demo video can be found at https://glowlts.github.io/.

Singing voice synthesis (SVS) systems are built to synthesize high-quality and expressive singing voice, in which the acoustic model generates the acoustic features (e.g., mel-spectrogram) given a music score. Previous singing acoustic models adopt a simple loss (e.g., L1 and L2) or generative adversarial network (GAN) to reconstruct the acoustic features, while they suffer from over-smoothing and unstable training issues respectively, which hinder the naturalness of synthesized singing. In this work, we propose DiffSinger, an acoustic model for SVS based on the diffusion probabilistic model. DiffSinger is a parameterized Markov chain that iteratively converts the noise into mel-spectrogram conditioned on the music score. By implicitly optimizing variational bound, DiffSinger can be stably trained and generate realistic outputs. To further improve the voice quality and speed up inference, we introduce a shallow diffusion mechanism to make better use of the prior knowledge learned by the simple loss. Specifically, DiffSinger starts generation at a shallow step smaller than the total number of diffusion steps, according to the intersection of the diffusion trajectories of the ground-truth mel-spectrogram and the one predicted by a simple mel-spectrogram decoder. Besides, we propose boundary prediction methods to locate the intersection and determine the shallow step adaptively. The evaluations conducted on a Chinese singing dataset demonstrate that DiffSinger outperforms state-of-the-art SVS work. Extensional experiments also prove the generalization of our methods on text-to-speech task (DiffSpeech). Audio samples: https://diffsinger.github.io. Codes: https://github.com/MoonInTheRiver/DiffSinger.

As a key component of talking face generation, lip movements generation determines the naturalness and coherence of the generated talking face video. Prior literature mainly focuses on speech-to-lip generation while there is a paucity in text-to-lip (T2L) generation. T2L is a challenging task and existing end-to-end works depend on the attention mechanism and autoregressive (AR) decoding manner. However, the AR decoding manner generates current lip frame conditioned on frames generated previously, which inherently hinders the inference speed, and also has a detrimental effect on the quality of generated lip frames due to error propagation. This encourages the research of parallel T2L generation. In this work, we propose a parallel decoding model for fast and high-fidelity text-to-lip generation (ParaLip). Specifically, we predict the duration of the encoded linguistic features and model the target lip frames conditioned on the encoded linguistic features with their duration in a non-autoregressive manner. Furthermore, we incorporate the structural similarity index loss and adversarial learning to improve perceptual quality of generated lip frames and alleviate the blurry prediction problem. Extensive experiments conducted on GRID and TCD-TIMIT datasets demonstrate the superiority of proposed methods.

Non-autoregressive text to speech (NAR-TTS) models have attracted much attention from both academia and industry due to their fast generation speed.One limitation of NAR-TTS models is that they ignore the correlation in time and frequency domains while generating speech mel-spectrograms, and thus cause blurry and over-smoothed results.In this work, we revisit this over-smoothing problem from a novel perspective: the degree of over-smoothness is determined by the gap between the complexity of data distributions and the capability of modeling methods.Both simplifying data distributions and improving modeling methods can alleviate the problem.Accordingly, we first study methods reducing the complexity of data distributions.Then we conduct a comprehensive study on NAR-TTS models that use some advanced modeling methods.Based on these studies, we find that 1) methods that provide additional condition inputs reduce the complexity of data distributions to model, thus alleviating the over-smoothing problem and achieving better voice quality.2) Among advanced modeling methods, Laplacian mixture loss performs well at modeling multimodal distributions and enjoys its simplicity, while GAN and Glow achieve the best voice quality while suffering from increased training or model complexity.3) The two categories of methods can be combined to further alleviate the over-smoothness and improve the voice quality.4) Our experiments on the multi-speaker dataset lead to similar conclusions as above and providing more variance information can reduce the difficulty of modeling the target data distribution and alleviate the requirements for model capacity.

Mengze Li, Tianbao Wang, Haoyu Zhang, Shengyu Zhang, Zhou Zhao, Jiaxu Miao, Wenqiao Zhang, Wenming Tan, Jin Wang, Peng Wang, Shiliang Pu, Fei Wu. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

We are interested in a novel task, singing voice beautification (SVB).Given the singing voice of an amateur singer, SVB aims to improve the intonation and vocal tone of the voice, while keeping the content and vocal timbre.Current automatic pitch correction techniques are immature, and most of them are restricted to intonation but ignore the overall aesthetic quality.Hence, we introduce Neural Singing Voice Beautifier (NSVB), the first generative model to solve the SVB task, which adopts a conditional variational autoencoder as the backbone and learns the latent representations of vocal tone.In NSVB, we propose a novel time-warping approach for pitch correction: Shape-Aware Dynamic Time Warping (SADTW), which ameliorates the robustness of existing time-warping approaches, to synchronize the amateur recording with the template pitch curve.Furthermore, we propose a latent-mapping algorithm in the latent space to convert the amateur vocal tone to the professional one.To achieve this, we also propose a new dataset containing parallel singing recordings of both amateur and professional versions.Extensive experiments on both Chinese and English songs demonstrate the effectiveness of our methods in terms of both objective and subjective metrics.

The performance of current Scene Graph Generation models is severely hampered by some hard-to-distinguish predicates, e.g., “woman-on/standing on/walking on-beach” or “woman-near/looking at/in front of-child”. While general SGG models are prone to predict head predicates and existing re-balancing strategies prefer tail categories, none of them can appropriately handle these hard-to-distinguish predicates. To tackle this issue, inspired by fine-grained image classification, which focuses on differentiating among hard-to-distinguish object classes, we propose a method named Fine-Grained Predicates Learning (FGPL) which aims at differentiating among hard-to-distinguish predicates for Scene Graph Generation task. Specifically, we first introduce a Predicate Lattice that helps SGG models to figure out fine-grained predicate pairs. Then, utilizing the Predicate Lattice, we propose a Category Discriminating Loss and an Entity Discriminating Loss, which both contribute to distinguishing fine-grained predicates while maintaining learned discriminatory power over recognizable ones. The proposed model-agnostic strategy significantly boosts the performances of three benchmark models (Transformer, VCTree, and Motif) by 22.8%, 24.1% and 21.7% of Mean Recall (mR@100) on the Predicate Classification sub-task, respectively. Our model also outperforms state-of-the-art methods by a large margin (i.e., 6.1%, 4.6%, and 3.2% of Mean Recall (mR@100)) on the Visual Genome dataset. Codes are publicly available11https://github.com/XinyuLyu/FGPL.

Audio-Guided video object segmentation is a challenging problem in visual analysis and editing, which automatically separates foreground objects from the background in a video sequence according to the referring audio expressions. However, existing referring video object segmentation works mainly focus on the guidance of text-based referring expressions, due to the lack of modeling the semantic representations of audio-video interaction contents. In this paper, we consider the problem of audio-guided video semantic segmentation from the viewpoint of end-to-end denoising encoder-decoder network learning. We propose the wavelet-based encoder network to learn the cross-modal representations of the video contents with audio-form queries. Specifically, we adopt the multi-head cross-modal attention layers to explore the potential relations of video and query contents. A 2-dimension discrete wavelet trans-form is merged into the transformer encoder to decompose the audio-video features. Next, we maximize mutual information between the encoded features and multi-modal features after cross-modal attention layers to enhance the au-dio guidance. Then, a self attention-free decoder network is developed to generate the target masks with frequency-domain transforms. In addition, we construct the first large-scale audio-guided video semantic segmentation dataset. The extensive experiments show the effectiveness of our method11Code is available at: https://github.com/asudahkzj/Wnet.git.

Audiovisual Event (AVE) localization requires the model to jointly localize an event by observing audio and visual information. However, in unconstrained videos, both information types may be inconsistent or suffer from severe background noise. Hence this paper proposes a novel cross-modal background suppression network for AVE task, operating at the time- and event-level, aiming to improve localization performance through suppressing asynchronous audiovisual background frames from the examined events and reducing redundant noise. Specifically, the time-level background suppression scheme forces the audio and visual modality to focus on the related information in the temporal dimension that the opposite modality considers essential, and reduces attention to the segments that the other modal considers as background. The event-level background suppression scheme uses the class activation sequences predicted by audio and visual modalities to control the final event category prediction, which can effectively sup-press noise events occurring accidentally in a single modality. Furthermore, we introduce a cross-modal gated attention scheme to extract relevant visual regions from complex scenes exploiting both global visual and audio signals. Extensive experiments show our method outperforms the state-of-the-art methods by a large margin in both supervised and weakly supervised AVE settings.11The source code and pre-trained models are publicly available at: https://github.com/marmot-xy/CMBS.

Most of the research to date focuses on bilingual sign language translation (BSLT). However, such models are in-efficient in building multilingual sign language translation systems. To solve this problem, we introduce the multilin-gual sign language translation (MSLT) task. It aims to use a single model to complete the translation between multiple sign languages and spoken languages. Then, we propose MLSLT, the first MSLT model, which contains two novel dy-namic routing mechanisms for controlling the degree ofpa-rameter sharing between different languages. Intra-layer language-specific routing controls the proportion of data flowing through shared parameters and language-specific parameters from the token level through a soft gate within the layer, and inter-layer language-specific routing controls and learns the data flow path of different languages at the language level through a soft gate between layers. In order to evaluate the performance of MLSLT, we collect the first publicly available multilingual sign language understanding dataset, Spreadthesign-Ten (SP-10), which contains up to 100 language pairs, e.g., CSL→en, GSG→zh. Experi-mental results show that the average performance of ML-SLT outperforms the baseline MSLT model and the com-bination of multiple BSLT models in many cases. In ad-dition, we also explore zero-shot translation in sign language and find that our model can achieve comparable performance to the supervised BSLT model on some language pairs. Dataset and more details are at https://mlslt.github.io/.

Expressive text-to-speech (TTS) has become a hot research topic recently, mainly focusing on modeling prosody in speech. Prosody modeling has several challenges: 1) the extracted pitch used in previous prosody modeling works have inevitable errors, which hurts the prosody modeling; 2) different attributes of prosody (e.g., pitch, duration and energy) are dependent on each other and produce the natural prosody together; and 3) due to high variability of prosody and the limited amount of high-quality data for TTS training, the distribution of prosody cannot be fully shaped. To tackle these issues, we propose ProsoSpeech, which enhances the prosody using quantized latent vectors pre-trained on large-scale unpaired and low-quality text and speech data. Specifically, we first introduce a word-level prosody encoder, which quantizes the low-frequency band of the speech and compresses prosody at-tributes in the latent prosody vector (LPV). Then we introduce an LPV predictor, which predicts LPV given word sequence. We pre-train the LPV predictor on large-scale text and low-quality speech data and fine-tune it on the high-quality TTS dataset. Finally, our model can generate expressive speech conditioned on the predicted LPV. Experimental results show that ProsoSpeech can generate speech with richer prosody compared with baseline methods.

Building a high-fidelity speech synthesis system with noisy speech data is a challenging but valuable task, which could significantly reduce the cost of data collection. Existing methods usually train speech synthesis systems based on the speech denoised with an enhancement model or feed noise information as a condition into the system. These methods certainly have some effect on inhibiting noise, but the quality and the prosody of their synthesized speech are still far away from natural speech. In this paper, we propose HiFiDenoise, a speech synthesis system with adversarial networks that can synthesize high-fidelity speech with low-quality and noisy speech data. Specifically, 1) to tackle the difficulty of noise modeling, we introduce multi-length adversarial training in the noise condition module. 2) To handle the problem of inaccurate pitch extraction caused by noise, we remove the pitch predictor in the acoustic model and also add discriminators on the mel-spectrogram generator. 3) In addition, we also apply HiFiDenoise to singing voice synthesis with a noisy singing dataset. Experiments show that our model outperforms the baseline by 0.36 and 0.44 in terms of MOS on speech and singing respectively.

Denoising diffusion probabilistic models (DDPMs) have recently achieved leading performances in many generative tasks. However, the inherited iterative sampling process costs hindered their applications to speech synthesis. This paper proposes FastDiff, a fast conditional diffusion model for high-quality speech synthesis. FastDiff employs a stack of time-aware location-variable convolutions of diverse receptive field patterns to efficiently model long-term time dependencies with adaptive conditions. A noise schedule predictor is also adopted to reduce the sampling steps without sacrificing the generation quality. Based on FastDiff, we design an end-to-end text-to-speech synthesizer, FastDiff-TTS, which generates high-fidelity speech waveforms without any intermediate feature (e.g., Mel-spectrogram). Our evaluation of FastDiff demonstrates the state-of-the-art results with higher-quality (MOS 4.28) speech samples. Also, FastDiff enables a sampling speed of 58x faster than real-time on a V100 GPU, making diffusion models practically applicable to speech synthesis deployment for the first time. We further show that FastDiff generalized well to the mel-spectrogram inversion of unseen speakers, and FastDiff-TTS outperformed other competing methods in end-to-end text-to-speech synthesis. Audio samples are available at https://FastDiff.github.io/.

The recent progress in non-autoregressive text-to-speech (NAR-TTS) has made fast and high-quality speech synthesis possible. However, current NAR-TTS models usually use phoneme sequence as input and thus cannot understand the tree-structured syntactic information of the input sequence, which hurts the prosody modeling. To this end, we propose SyntaSpeech, a syntax-aware and light-weight NAR-TTS model, which integrates tree-structured syntactic information into the prosody modeling modules in PortaSpeech. Specifically, 1) We build a syntactic graph based on the dependency tree of the input sentence, then process the text encoding with a syntactic graph encoder to extract the syntactic information. 2) We incorporate the extracted syntactic encoding with PortaSpeech to improve the prosody prediction. 3) We introduce a multi-length discriminator to replace the flow-based post-net in PortaSpeech, which simplifies the training pipeline and improves the inference speed, while keeping the naturalness of the generated audio. Experiments on three datasets not only show that the tree-structured syntactic information grants SyntaSpeech the ability to synthesize better audio with expressive prosody, but also demonstrate the generalization ability of SyntaSpeech to adapt to multiple languages and multi-speaker text-to-speech. Ablation studies demonstrate the necessity of each component in SyntaSpeech. Source code and audio samples are available at https://syntaspeech.github.io.

Zero-shot text-based singing editing enables singing voice modification based on the given edited lyrics without any additional data from the target singer. However, due to the different demands, challenges occur when applying existing speech editing methods to singing voice editing task, mainly including the lack of systematic consideration concerning prosody in insertion and deletion, as well as the trade-off between the naturalness of pronunciation and the preservation of prosody in replacement. In this paper we propose EditSinger, which is a novel singing voice editing model with specially designed diverse prosody modules to overcome the challenges above. Specifically, 1) a general masked variance adaptor is introduced for the comprehensive prosody modeling of the inserted lyrics and the transition of deletion boundary; and 2) we further design a fusion pitch predictor for replacement. By disentangling the reference pitch and fusing the predicted pronunciation, the edited pitch can be reconstructed, which could ensure a natural pronunciation while preserving the prosody of the original audio. In addition, to the best of our knowledge, it is the first zero-shot text-based singing voice editing system. Our experiments conducted on the OpenSinger prove that EditSinger can synthesize high-quality edited singing voices with natural prosody according to the corresponding operations.

One of the challenging factors in real application of sign language translation (SLT) is inter-signer variation. With the assumption that the pre-trained translation model cannot cover all the signers, the adaptation capability for unseen signers is of great concern. In this paper, we take a completely different perspective for SLT, called signer-adaptive SLT, which mainly considers the transferable ability of SLT systems. To attack this challenging problem, we propose MC-SLT, a novel meta-learning framework that could exploit additional new-signer data via a support set, and output a signer-adaptive model via a few-gradient-step update. Considering the various degrees of style discrepancies of different words performed by multiple signers, we further devise diversity-aware meta-adaptive weights for the token-wise cross-entropy losses. Besides, to improve the training robustness, we adopt the self-guided curriculum learning scheme that first captures the global curricula from each signer to avoid falling into a bad local optimum early, and then learns the curricula of individualities to improve the model adaptability for learning signer-specific knowledge. We re-construct the existing standard datasets of SLT for the signer-adaptive setting and establish a new benchmark for subsequent research.

Deep generative models have achieved significant progress in speech synthesis to date, while high-fidelity singing voice synthesis is still an open problem for its long continuous pronunciation, rich high-frequency parts, and strong expressiveness. Existing neural vocoders designed for text-to-speech cannot directly be applied to singing voice synthesis because they result in glitches and poor high-frequency reconstruction. In this work, we propose SingGAN, a generative adversarial network designed for high-fidelity singing voice synthesis. Specifically, 1) to alleviate the glitch problem in the generated samples, we propose source excitation with the adaptive feature learning filters to expand the receptive field patterns and stabilize long continuous signal generation; and 2) SingGAN introduces global and local discriminators at different scales to enrich low-frequency details and promote high-frequency reconstruction; and 3) To improve the training efficiency, SingGAN includes auxiliary spectrogram losses and sub-band feature matching penalty loss. To the best of our knowledge, SingGAN is the first work designed toward high-fidelity singing voice vocoding. Our evaluation of SingGAN demonstrates the state-of-the-art results with higher-quality (MOS 4.05) samples. Also, SingGAN enables a sample speed of 50x faster than real-time on a single NVIDIA 2080Ti GPU. We further show that SingGAN generalizes well to the mel-spectrogram inversion of unseen singers, and the end-to-end singing voice synthesis system SingGAN-SVS enjoys a two-stage pipeline to transform the music scores into expressive singing voices.

Sign Language Production (SLP) aims to translate a spoken language description to its corresponding continuous sign language sequence. A prevailing solution for this problem is in a two-staged manner: it formulates SLP as two sub-tasks, i.e., Text to Gloss (T2G) translation and Gloss to Pose (G2P) animation, with gloss annotations as pivots. Although two-staged approaches achieve better performance than their direct translation counterparts, the requirement of gloss intermediaries causes a parallel data bottleneck. In this paper, to reduce reliance on gloss annotations in two-staged approaches, we propose DualSign, a semi-supervised two-staged SLP framework, which can effectively utilize partially gloss-annotated text-pose pairs and monolingual gloss data. The key component of DualSign is a novel Balanced Multi-Modal Multi-Task Dual Transformation (BM3T-DT) method, where two well-designed models, i.e., a Multi-Modal T2G model (MM-T2G) and a Multi-Task G2P model (MT-G2P), are jointly trained by leveraging their task duality and unlabeled data. After applying BM3T-DT, we derive the expected uni-modal T2G model from the well-trained MM-T2G with knowledge distillation. Considering that the MM-T2G may suffer from modality imbalance when decoding with multiple input modalities, we devise a cross-modal balancing loss, further boosting the system's overall performance. Extensive experiments conducted on the PHOENIX14T dataset show the effectiveness of our approach in the semi-supervised setting. By training with additionally collected unlabeled data, DualSign substantially improves previous state-of-the-art SLP methods.

Denoising diffusion probabilistic models (DDPMs) have recently achieved leading performances in many generative tasks. However, the inherited iterative sampling process costs hinder their applications to text-to-speech deployment. Through the preliminary study on diffusion model parameterization, we find that previous gradient-based TTS models require hundreds or thousands of iterations to guarantee high sample quality, which poses a challenge for accelerating sampling. In this work, we propose ProDiff, on progressive fast diffusion model for high-quality text-to-speech. Unlike previous work estimating the gradient for data density, ProDiff parameterizes the denoising model by directly predicting clean data to avoid distinct quality degradation in accelerating sampling. To tackle the model convergence challenge with decreased diffusion iterations, ProDiff reduces the data variance in the target site via knowledge distillation. Specifically, the denoising model uses the generated mel-spectrogram from an N-step DDIM teacher as the training target and distills the behavior into a new model with N/2 steps. As such, it allows the TTS model to make sharp predictions and further reduces the sampling time by orders of magnitude. Our evaluation demonstrates that ProDiff needs only 2 iterations to synthesize high-fidelity mel-spectrograms, while it maintains sample quality and diversity competitive with state-of-the-art models using hundreds of steps. ProDiff enables a sampling speed of 24x faster than real-time on a single NVIDIA 2080Ti GPU, making diffusion models practically applicable to text-to-speech synthesis deployment for the first time. Our extensive ablation studies demonstrate that each design in ProDiff is effective, and we further show that ProDiff can be easily extended to the multi-speaker setting.

Fingerspelling detection, which aims to localize and recognize fingerspelling gestures in raw, untrimmed videos, is a nascent but important research area that could help bridge the communication gap between deaf people and others. Many existing works tend to exploit additional knowledge, such as pose annotations, and newly datasets for performance improvement. However, in real-world applications, additional data collection and annotation require tremendous human efforts that are not always affordable. In this paper, we propose the Weakly-supervised Disentanglement Network, namely WED, that requires no additional knowledge, and better exploits the video-sentence weak supervisions. Specifically, WED incorporates two critical components: 1) Masked Disentanglement Module, which employs a Variational Autoencoder for signed letters disentanglement. Each latent factor in the VAE corresponds to a particular signed letter, and we mask latent factors corresponding to letters that do not appear in the video during decoding. Compared to the vanilla VAE, the masked reconstruction leverages the video-sentence weak supervision, leading to a better sign language oriented disentanglement; and 2) the Dynamic Memory Network module, which leverages the disentangled sign knowledge as prior knowledge and reference for sign-related frame identification and gesture recognition through a carefully designed memory reading component. We conduct extensive experiments on the benchmark ChicagoFSWild and ChicagoFSWild+ datasets. Empirical studies validate that the WED network achieves effective sign gesture disentanglement, contributing to the state-of-the-art performance for fingerspelling detection and recognition.

Video Object Grounding (VOG) is the problem of associating spatial object regions in the video to a descriptive natural language query. This is a challenging vision-language task that necessitates constructing the correct cross-modal correspondence and modeling the appropriate spatio-temporal context of the query video and caption, thereby localizing the specific objects accurately. In this paper, we tackle this task by a novel framework called HiErarchical spatio-tempoRal reasOning (HERO) with contrastive action correspondence. We study the VOG task at two aspects that prior works overlooked: (1) Contrastive Action Correspondence-aware Retrieval. Notice that the fine-grained video semantics (e.g., multiple actions) is not totally aligned with the annotated language query (e.g., single action), we first introduce the weakly-supervised contrastive learning that classifies the video as action-consistent and action-independent frames relying on the video-caption action semantic correspondence. Such a design can build the fine-grained cross-modal correspondence for more accurate subsequent VOG. (2) Hierarchical Spatio-temporal Modeling Improvement. While transformer-based VOG models present their potential in sequential modality (i.e., video and caption) modeling, existing evidence also indicates that the transformer suffers from the issue of the insensitive spatio-temporal locality. Motivated by that, we carefully design the hierarchical reasoning layers to decouple fully connected multi-head attention and remove the redundant interfering correlations. Furthermore, our proposed pyramid and shifted alignment mechanisms are effective to improve the cross-modal information utilization of neighborhood spatial regions and temporal frames. We conducted extensive experiments to show our HERO outperforms existing techniques by achieving significant improvement on two benchmark datasets.

Set-based face recognition (SFR) aims to recognize the face sets in the unconstrained scenario, where the appearance of same identity may change dramatically with extreme variances (e.g., illumination, pose, expression). We argue that the two crucial issues in SFR, the face quality and burstiness, are both identity-irrelevant and variance-relevant. The quality and burstiness assessment are interfered with by the entanglement of identity, and the face recognition is interfered with by the entanglement of variance. Thus we propose to separate the identity features with the variance features in a light-weighted set-based disentanglement framework.

Unconstrained lip-to-speech synthesis aims to generate corresponding speeches from silent videos of talking faces with no restriction on head poses or vocabulary. Current works mainly use sequence-to-sequence models to solve this problem, either in an autoregressive architecture or a flow-based non-autoregressive architecture. However, these models suffer from several drawbacks: 1) Instead of directly generating audios, they use a two-stage pipeline that first generates mel-spectrograms and then reconstructs audios from the spectrograms. This causes cumbersome deployment and degradation of speech quality due to error propagation; 2) The audio reconstruction algorithm used by these models limits the inference speed and audio quality, while neural vocoders are not available for these models since their output spectrograms are not accurate enough; 3) The autoregressive model suffers from high inference latency, while the flow-based model has high memory occupancy: neither of them is efficient enough in both time and memory usage. To tackle these problems, we propose FastLTS, a non-autoregressive end-to-end model which can directly synthesize high-quality speech audios from unconstrained talking videos with low latency, and has a relatively small model size. Besides, different from the widely used 3D-CNN visual frontend for lip movement encoding, we for the first time propose a transformer-based visual frontend for this task. Experiments show that our model achieves 19.76x speedup for audio waveform generation compared with the current autoregressive model on input sequences of 3 seconds, and obtains superior audio quality.

In this paper, we introduce a new task, spoken video grounding (SVG), which aims to localize the desired video fragments from spoken language descriptions. Compared with using text, employing audio requires the model to directly exploit the useful phonemes and syllables related to the video from raw speech. Moreover, we randomly add environmental noises to this speech audio, further increasing the difficulty of this task and better simulating real applications. To rectify the discriminative phonemes and extract video-related information from noisy audio, we develop a novel video-guided curriculum learning (VGCL) during the audio pre-training process, which can make use of the vital visual perceptions to help understand the spoken language and suppress the external noise. Considering during inference the model can not obtain ground truth video segments, we design a curriculum strategy that gradually shifts the input video from the ground truth to the entire video content during pre-training. Finally, the model can learn how to extract critical visual information from the entire video clip to help understand the spoken language. In addition, we collect the first large-scale spoken video grounding dataset based on ActivityNet, which is named as ActivityNet Speech dataset. Extensive experiments demonstrate our proposed video-guided curriculum learning can facilitate the pre-training process to obtain a mutual audio encoder, significantly promoting the performance of spoken video grounding tasks. Moreover, we prove that in the case of noisy sound, our model outperforms the method that grounding video with ASR transcripts, further demonstrating the effectiveness of our curriculum strategy.

Style transfer for out-of-domain (OOD) speech synthesis aims to generate speech samples with unseen style (e.g., speaker identity, emotion, and prosody) derived from an acoustic reference, while facing the following challenges: 1) The highly dynamic style features in expressive voice are difficult to model and transfer; and 2) the TTS models should be robust enough to handle diverse OOD conditions that differ from the source data. This paper proposes GenerSpeech, a text-to-speech model towards high-fidelity zero-shot style transfer of OOD custom voice. GenerSpeech decomposes the speech variation into the style-agnostic and style-specific parts by introducing two components: 1) a multi-level style adaptor to efficiently model a large range of style conditions, including global speaker and emotion characteristics, and the local (utterance, phoneme, and word-level) fine-grained prosodic representations; and 2) a generalizable content adaptor with Mix-Style Layer Normalization to eliminate style information in the linguistic content representation and thus improve model generalization. Our evaluations on zero-shot style transfer demonstrate that GenerSpeech surpasses the state-of-the-art models in terms of audio quality and style similarity. The extension studies to adaptive style transfer further show that GenerSpeech performs robustly in the few-shot data setting. Audio samples are available at \url{https://GenerSpeech.github.io/}.

Polyphone disambiguation aims to capture accurate pronunciation knowledge from natural text sequences for reliable Text-to-speech (TTS) systems. However, previous approaches require substantial annotated training data and additional efforts from language experts, making it difficult to extend high-quality neural TTS systems to out-of-domain daily conversations and countless languages worldwide. This paper tackles the polyphone disambiguation problem from a concise and novel perspective: we propose Dict-TTS, a semantic-aware generative text-to-speech model with an online website dictionary (the existing prior information in the natural language). Specifically, we design a semantics-to-pronunciation attention (S2PA) module to match the semantic patterns between the input text sequence and the prior semantics in the dictionary and obtain the corresponding pronunciations; The S2PA module can be easily trained with the end-to-end TTS model without any annotated phoneme labels. Experimental results in three languages show that our model outperforms several strong baseline models in terms of pronunciation accuracy and improves the prosody modeling of TTS systems. Further extensive analyses demonstrate that each design in Dict-TTS is effective. The code is available at https://github.com/Zain-Jiang/Dict-TTS.

The lack of publicly available high-quality and accurately labeled datasets has long been a major bottleneck for singing voice synthesis (SVS). To tackle this problem, we present M4Singer, a free-to-use Multi-style, Multi-singer Mandarin singing collection with elaborately annotated Musical scores as well as its benchmarks. Specifically, 1) we construct and release a large high-quality Chinese singing voice corpus, which is recorded by 20 professional singers, covering 700 Chinese pop songs as well as all the four SATB types (i.e., soprano, alto, tenor, and bass); 2) we take extensive efforts to manually compose the musical scores for each recorded song, which are necessary to the study of the prosody modeling for SVS. 3) To facilitate the use and demonstrate the quality of M4Singer, we conduct four different benchmark experiments: score-based SVS, controllable singing voice (CSV), singing voice conversion (SVC) and automatic music transcription (AMT).

Diffusion Probabilistic Models (DPMs) have shown a powerful capacity of generating high-quality image samples. Recently, diffusion autoencoders (Diff-AE) have been proposed to explore DPMs for representation learning via autoencoding. Their key idea is to jointly train an encoder for discovering meaningful representations from images and a conditional DPM as the decoder for reconstructing images. Considering that training DPMs from scratch will take a long time and there have existed numerous pre-trained DPMs, we propose \textbf{P}re-trained \textbf{D}PM \textbf{A}uto\textbf{E}ncoding (\textbf{PDAE}), a general method to adapt existing pre-trained DPMs to the decoders for image reconstruction, with better training efficiency and performance than Diff-AE. Specifically, we find that the reason that pre-trained DPMs fail to reconstruct an image from its latent variables is due to the information loss of forward process, which causes a gap between their predicted posterior mean and the true one. From this perspective, the classifier-guided sampling method can be explained as computing an extra mean shift to fill the gap, reconstructing the lost class information in samples. These imply that the gap corresponds to the lost information of the image, and we can reconstruct the image by filling the gap. Drawing inspiration from this, we employ a trainable model to predict a mean shift according to encoded representation and train it to fill as much gap as possible, in this way, the encoder is forced to learn as much information as possible from images to help the filling. By reusing a part of network of pre-trained DPMs and redesigning the weighting scheme of diffusion loss, PDAE can learn meaningful representations from images efficiently. Extensive experiments demonstrate the effectiveness, efficiency and flexibility of PDAE.

Aiming to locate the object that emits a specified sound in complex scenes, the task of sounding object localization bridges two perception-oriented modalities of vision and acoustics, and brings enormous research value to the comprehensive perceptual understanding of machine intelligence. Although there are massive training data collected in this field, few of them contain accurate bounding box annotations, hindering the learning process and further application of proposed models. In order to address this problem, we try to explore an effective multi-modal knowledge transfer strategy to obtain precise knowledge from other similar tasks and transfer it through well-aligned multi-modal data to deal with this task in a zero-resource manner. Concretely, we design and propose a novel \textit{Two-stream Universal Referring localization Network} (TURN), which is composed of a localization stream and an alignment stream to carry out different functions. The former is utilized to extract the knowledge related to referring object localization from the image grounding task, while the latter is devised to learn a universal semantic space shared between texts and audios. Moreover, we further develop an adaptive sampling strategy to automatically identify the overlap between different data domains, thus boosting the performance and stability of our model. The extensive experiments on various publicly-available benchmarks demonstrate that TURN can achieve competitive performance compared with the state-of-the-art approaches without using any data in this field, which verifies the feasibility of our proposed mechanisms and strategies.

In recent years, online short videos have become more popular, especially as an online advertising intermediary. To better understand their effects as advertisements, it is essential to analyze the causal relations of online short videos on consumer behaviors. Our study is based on fine-grained consumer behavior data from a world-leading e-commerce platform, i.e., Taobao.com. We first decompose the total causal effects into informative effects and persuasive effects following a common practice in the economic literature. Moreover, we extract the subjectivity scores of short videos through a dictionary-based subjectivity analysis model and evaluate the correlation between the subjectivity scores and each causal effect. The findings of this paper are as follows: First, both causal effects (i.e., informative and persuasive effects) are significant. Second, these effects have a strong correlation with the short videos' subjectivity scores. Third, the signs of these correlations vary with the prices of the products. Our results not only shed light on the research of how short videos exert influence on online consumers, but also give sellers advice on better video design and recommendation.

Self-supervised learning, especially contrastive learning, has made an outstanding contribution to the development of many deep learning research fields. Recently, researchers in the acoustic signal processing field noticed its success and leveraged contrastive learning for better music representation. Typically, existing approaches maximize the similarity between two distorted audio segments sampled from the same music. In other words, they ensure a semantic agreement at the music level. However, those coarse-grained methods neglect some inessential or noisy elements at the frame level, which may be detrimental to the model to learn the effective representation of music. Towards this end, this paper proposes a novel Positive-nEgative frame mask for Music Representation based on the contrastive learning framework, abbreviated as PEMR. Concretely, PEMR incorporates a Positive-Negative Mask Generation module, which leverages transformer blocks to generate frame masks on Log-Mel spectrogram. We can generate self-augmented negative and positive samples by masking important components or inessential components, respectively. We devise a novel contrastive learning objective to accommodate both self-augmented positives/negatives sampled from the same music. We conduct experiments on four public datasets. The experimental results of two music-related downstream tasks, music classification and cover song identification, demonstrate the generalization ability and transferability of music representation learned by PEMR.

Effectively representing users lie at the core of modern recommender systems. Since users’ interests naturally exhibit multiple aspects, it is of increasing interest to develop multi-interest frameworks for recommendation, rather than represent each user with an overall embedding. Despite their effectiveness, existing methods solely exploit the encoder (the forward flow) to represent multiple aspects of interests. However, without explicit regularization, the interest embeddings may not be distinct from each other nor semantically reflect representative historical items. Towards this end, we propose the Re4 framework, which leverages the backward flow to reexamine each interest embedding. Specifically, Re4 encapsulates three backward flows, i.e., 1) Re-contrast, which drives each interest embedding to be distinct from other interests using contrastive learning; 2) Re-attend, which ensures the interest-item correlation estimation in the forward flow to be consistent with the criterion used in final recommendation; and 3) Re-construct, which ensures that each interest embedding can semantically reflect the information of representative items that relate to the corresponding interest. We demonstrate the novel forward-backward multi-interest paradigm on ComiRec, and perform extensive experiments on three real-world datasets. Empirical studies validate that Re4 helps to learn learning distinct and effective multi-interest representations.

As a task of cross-modal retrieval between long videos and paragraphs, video-paragraph retrieval is a non-trivial task. Unlike traditional video-text retrieval, the video in video-paragraph retrieval usually contains multiple clips. Each clip corresponds to a descriptive sentence; all the sentences constitute the corresponding paragraph of the video. Previous methods for video-paragraph retrieval usually encode videos and para-graphs from segment-level (clips and sentences) and overall-level (videos and paragraphs). However, there are also contents about actions and objects that exist in the segment. Hence, we propose a Local-Global Graph Pooling Network (LGGP) via Mutual Information Maximization for video-paragraph retrieval. Our model disentangles videos and paragraphs into four levels: overall-level, segment-level, motion-level, and object-level. We construct the Hierarchical Local Graph (segment-level, motion-level, and object-level) and the Hierarchical Global Graph (overall-level, segment-level, motion-level, and object-level), respectively, for semantic interaction among different levels. Meanwhile, to obtain hierarchical pooling features with fine-grained semantic information, we design hierarchical graph pooling methods to maximize the mutual information between pooling features and corresponding graph nodes. We evaluate our model on two video-paragraph retrieval datasets with three different video features. The experimental results show that our model establishes state-of-the-art results for video-paragraph retrieval. Our code will be released athttps://github.com/PengchengZhang1997/LGGP.

In e-commerce, product related video is important content to introduce product characteristics and attract consumers. Especially in the recommendation system of e-commerce platform, video highlight detection methods are usually adopted to capture the most attractive clips for showing to consumers, so as to improve the click through rate of products. However, the effect of the current research methods applied to the actual scene is not satisfactory. Compared with other video understanding tasks, video highlight detection is relatively abstract and subjective, and it is difficult to make accurate judgment only by using visual information. Consequently, we put forward multi-modal video highlight detection task, which introduces video related linguistic information as supervised information. And we propose a graph-based commodity-aware model to solve multi-modal video highlight detection in e-commerce scene. Our model consists of multi-modal highlight detection stage and graph-based fine-tuning stage, in which we adopt graph aggregation method to fuse multi-source natural language information and introduce effective visual feature composition method for graph convolution network based highlight detection. Besides, we release the largest e-commerce video highlight detection dataset, TaoHighlight, in which the videos and related data are collected from Taobao e-commerce platform. Our model achieves state-of-art in all separate categories and overall dataset of TaoHighlight, which shows the superiority of our model.

Face aging and rejuvenation is applied to predict what a person looks like at different ages. While prior work brought about a significant progress in this topic, there are two central problems remaining to be solved : 1) most prior works require sequential data during training, while it is very rare in existing datasets; and 2) how to render an aging face and preserve personality at the same time. To deal with these problems, we develop a novel dual conditional GANs mechanism, thus aging faces can be trained with multiple sets of unlabeled facial images of different ages. Our basic architecture is AgeGAN, in which the primal conditional GAN converts input faces to other ages based on relevant age conditions, and the dual conditional GAN learns to invert the task. We further improve our networks, termed AgeGAN++, in which we share the weights between the primal part and the dual part to to streamline the model. Moreover, in order to get more sensible results, a representation disentanglement component is integrated with the latent facial representation, and an enhanced discriminator is applied on the generated process. In addition, we firstly perform an interpolation experiment to demonstrate that our generators are powerful and effective for face aging and rejuvenation. Experimental results on four public datasets demonstrate the appealing performance of the proposed methods by comparing with the state-of-the-art methods. Our code and a demo are released athttps://github.com/Sherry-JQ/AgeGAN.

Code summarization (aka comment generation) provides a high-level natural language description of the function performed by code, which can benefit the software maintenance, code categorization and retrieval. To the best of our knowledge, the state-of-the-art approaches follow an encoder-decoder framework which encodes source code into a hidden space and later decodes it into a natural language space. Such approaches suffer from the following drawbacks: (a) they are mainly input by representing code as a sequence of tokens while ignoring code hierarchy; (b) most of the encoders only input simple features (e.g., tokens) while ignoring the features that can help capture the correlations between comments and code; (c) the decoders are typically trained to predict subsequent words by maximizing the likelihood of subsequent ground truth words, while in real world, they are excepted to generate the entire word sequence from scratch. As a result, such drawbacks lead to inferior and inconsistent comment generation accuracy. To address the above limitations, this paper presents a new code summarization approach using hierarchical attention network by incorporating multiple code features, including type-augmented abstract syntax trees and program control flows. Such features, along with plain code sequences, are injected into a deep reinforcement learning (DRL) framework (e.g., actor-critic network) for comment generation. Our approach assigns weights (pays “attention”) to tokens and statements when constructing the code representation to reflect the hierarchical code structure under different contexts regarding code features (e.g., control flows and abstract syntax trees). Our reinforcement learning mechanism further strengthens the prediction results through the actor network and the critic network, where the actor network provides the confidence of predicting subsequent words based on the current state, and the critic network computes the reward values of all the possible extensions of the current state to provide global guidance for explorations. Eventually, we employ an advantage reward to train both networks and conduct a set of experiments on a real-world dataset. The experimental results demonstrate that our approach outperforms the baselines by around 22 to 45 percent in BLEU-1 and outperforms the state-of-the-art approaches by around 5 to 60 percent in terms of S-BLEU and C-BLEU.

Personalized recommendation system has become pervasive in various video platform.Many effective methods have been proposed, but most of them didn’t capture the user’s multilevel interest trait and dependencies between their viewed micro-videos well. To solve these problems, we propose a Self-over-Co Attention module to enhance user’s interest representation. In particular, we first use co-attention to model correlation patterns across different levels and then use self attention to modelcorrelation patterns within a specific level. Experimental results on filtered public datasets verify that our presented module is useful.

Given an untrimmed video and a query sentence, cross-modal video moment retrieval aims to rank a video moment from pre-segmented video moment candidates that best matches the query sentence. Pioneering work typically learns the representations of the textual and visual content separately and then obtains the interactions or alignments between different modalities. However, the task of cross-modal video moment retrieval is not yet thoroughly addressed as it needs to further identify the fine-grained differences of video moment candidates with high repeatability and similarity. Moveover, the relation among objects in both video and sentence is intuitive and efficient for understanding semantics but is rarely considered.Toward this end, we contribute a multi-modal relational graph to capture the interactions among objects from the visual and textual content to identify the differences among similar video moment candidates. Specifically, we first introduce a visual relational graph and a textual relational graph to form relation-aware representations via message propagation. Thereafter, a multi-task pre-training is designed to capture domain-specific knowledge about objects and relations, enhancing the structured visual representation after explicitly defined relation. Finally, the graph matching and boundary regression are employed to perform the cross-modal retrieval. We conduct extensive experiments on two datasets about daily activities and cooking activities, demonstrating significant improvements over state-of-the-art solutions.

Temporal video grounding aims to localize the target segment which is semantically aligned with the given sentence in an untrimmed video. Existing methods can be divided into two main categories, including proposal-based approaches and proposal-free approaches. However, the former ones suffer from the extra cost of generating proposals and inflexibility in determining fine-grained boundaries, and the latter ones usually attempt to decide the start and end timestamps directly, which brings about much difficulty and inaccuracy. In this paper, we convert this task into a multi-step decision problem and propose a novel Cascaded Prediction Network (CPN) to generate the grounding result in a coarse-to-fine manner. Concretely, we first encode video and query into the same latent space and fuse them into integrated representations. Afterwards, we construct a segment-tree-based structure and make predictions via decision navigation and signal decomposition in a cascaded way. We evaluate our proposed method on three large-scale publicly available benchmarks, namely ActivityNet Caption, Charades-STA and TACoS, where our CPN surpasses the performance of the state-of-the-art methods.

Existing reasoning tasks often have an important assumption that the input contents can be always accessed while reasoning, requiring unlimited storage resources and suffering from severe time delay on long sequences. To achieve efficient reasoning on long sequences with limited storage resources, memory augmented neural networks introduce a human-like write-read memory to compress and memorize the long input sequence in one pass, trying to answer subsequent queries only based on the memory. But they have two serious drawbacks: 1) they continually update the memory from current information and inevitably forget the early contents; 2) they do not distinguish what information is important and treat all contents equally. In this paper, we propose the Rehearsal Memory (RM) to enhance long-sequence memorization by self-supervised rehearsal with a history sampler. To alleviate the gradual forgetting of early information, we design self-supervised rehearsal training with recollection and familiarity tasks. Further, we design a history sampler to select informative fragments for rehearsal training, making the memory focus on the crucial information. We evaluate the performance of our rehearsal memory by the synthetic bAbI task and several downstream tasks, including text/video question answering and recommendation on long sequences.

Federated learning enables collaborative training of machine learning models under strict privacy restrictions and federated text-to-speech aims to synthesize natural speech of multiple users with a few audio training samples stored in their devices locally. However, federated text-to-speech faces several challenges: very few training samples from each speaker are available, training samples are all stored in local device of each user, and global model is vulnerable to various attacks. In this paper, we propose a novel federated learning architecture based on continual learning approaches to overcome the difficulties above. Specifically, 1) we use gradual pruning masks to isolate parameters for preserving speakers' tones; 2) we apply selective masks for effectively reusing knowledge from tasks; 3) a private speaker embedding is introduced to keep users' privacy. Experiments on a reduced VCTK dataset demonstrate the effectiveness of FedSpeech: it nearly matches multi-task training in terms of multi-speaker speech quality; moreover, it sufficiently retains the speakers' tones and even outperforms the multi-task training in the speaker similarity experiment.

Recently, there has been an increasing interest in neural speech synthesis. While the deep neural network achieves the state-of-the-art result in text-to-speech (TTS) tasks, how to generate a more emotional and more expressive speech is becoming a new challenge to researchers due to the scarcity of high-quality emotion speech dataset and the lack of advanced emotional TTS model. In this paper, we first briefly introduce and publicly release a Mandarin emotion speech dataset including 9,724 samples with audio files and its emotion human-labeled annotation. After that, we propose a simple but efficient architecture for emotional speech synthesis called EMSpeech. Unlike those models which need additional reference audio as input, our model could predict emotion labels just from the input text and generate more expressive speech conditioned on the emotion embedding. In the experiment phase, we first validate the effectiveness of our dataset by an emotion classification task. Then we train our model on the proposed dataset and conduct a series of subjective evaluations. Finally, by showing a comparable performance in the emotional speech synthesis task, we successfully demonstrate the ability of the proposed model.

Audio super-resolution is the task of constructing a highresolution (HR) audio from a low-resolution (LR) audio by adding the missing band.Previous methods based on convolutional neural networks and mean squared error training objective have relatively low performance, while adversarial generative models are difficult to train and tune.Recently, normalizing flow has attracted a lot of attention for its high performance, simple training and fast inference.In this paper, we propose WSRGlow, a Glow-based waveform generative model to perform audio super-resolution.Specifically, 1) we integrate WaveNet and Glow to directly maximize the exact likelihood of the target HR audio conditioned on LR information; and 2) to exploit the audio information from low-resolution audio, we propose an LR audio encoder and an STFT encoder, which encode the LR information from the time domain and frequency domain respectively.The experimental results show that the proposed model is easier to train and outperforms the previous works in terms of both objective and perceptual quality.WSRGlow is also the first model to produce 48kHz waveforms from 12kHz LR audio.Audio samples are available at https://zkx06111.github.io/wsrglow/.

High-fidelity multi-singer singing voice synthesis is challenging for neural vocoder due to the singing voice data shortage, limited singer generalization, and large computational cost. Existing open corpora could not meet requirements for high-fidelity singing voice synthesis because of the scale and quality weaknesses. Previous vocoders have difficulty in multi-singer modeling, and a distinct degradation emerges when conducting unseen singer singing voice generation. To accelerate singing voice researches in the community, we release a large-scale, multi-singer Chinese singing voice dataset OpenSinger. To tackle the difficulty in unseen singer modeling, we propose Multi-Singer, a fast multi-singer vocoder with generative adversarial networks. Specifically, 1) Multi-Singer uses a multi-band generator to speed up both training and inference procedure. 2) to capture and rebuild singer identity from the acoustic feature (i.e., mel-spectrogram), Multi-Singer adopts a singer conditional discriminator and conditional adversarial training objective. 3) to supervise the reconstruction of singer identity in the spectrum envelopes in frequency domain, we propose an auxiliary singer perceptual loss. The joint training approach effectively works in GANs for multi-singer voices modeling. Experimental results verify the effectiveness of OpenSinger and show that Multi-Singer improves unseen singer singing voices modeling in both speed and quality over previous methods. The further experiment proves that combined with FastSpeech 2 as the acoustic model, Multi-Singer achieves strong robustness in the multi-singer singing voice synthesis pipeline.

Sign Language Production (SLP) aims to automatically translate a spoken language description to its corresponding sign language video. The core procedure of SLP is to transform sign gloss intermediaries into sign pose sequences (G2P). Most existing methods for G2P are based on sequential autoregression or sequence-to-sequence encoder-decoder learning. However, by generating target pose frames conditioned on the previously generated ones, these models are prone to bringing issues such as error accumulation and high inference latency. In this paper, we argue that such issues are mainly caused by adopting autoregressive manner. Hence, we propose a novel Non-AuToregressive (NAT) model with a parallel decoding scheme, as well as an External Aligner for sequence alignment learning. Specifically, we extract alignments from the external aligner by monotonic alignment search for gloss duration prediction, which is used by a length regulator to expand the source gloss sequence to match the length of the target sign pose sequence for parallel sign pose generation. Furthermore, we devise a spatial-temporal graph convolutional pose generator in the NAT model to generate smoother and more natural sign pose sequences. Extensive experiments conducted on PHOENIX14T dataset show that our proposed model outperforms state-of-the-art autoregressive models in terms of speed and quality.

Sign language translation aims at directly translating a sign language video into a natural sentence. The majority of existing methods take the video-sentence pairs labeled by multiple specific signers as training and testing samples. However, such setting does not fit in with the real-world applications. A practicable sign language translation system is supposed to provide accurate translation results for unseen signers. In this paper, we mainly attack the signer-independent setting and focus on augmenting the generalization ability of translation model. To adapt to the challenging setting, we propose a novel framework called contrastive disentangled meta-learning (CDM), which develops several improvements in both deep architecture and training mode. Specifically, based on the minimax entropy objective, a disentangled module with adaptive gated units is developed to decouple the signer-specific and task-specific representation in the encoder. Besides, we facilitate the frame-word alignments by leveraging contrastive constraints between the obtained task-specific representation and the decoding output. The disentangled and contrastive modules could provide complementary information for each other. As for the training mode, we encourage the model to perform well in the simulated signer-independent scenarios by finding the generalized learning directions in the meta-learning process. Considering that vanilla meta-learning methods utilize the multiple specific signers insufficiently, we adopt a fine-grained learning strategy that simultaneously conducts meta-learning in a variety of domain shift scenarios in each iteration. Extensive experiments on the benchmark dataset RWTH-PHOENIX-Weather-2014T(PHOENIX14T) show that CDM could achieve competitive results compared with the state-of-the-art methods.

Lip reading, aiming to recognize spoken sentences according to the given video of lip movements without relying on the audio stream, has attracted great interest due to its application in many scenarios. Although prior works that explore lip reading have obtained salient achievements, they are all trained in a non-simultaneous manner where the predictions are generated requiring access to the full video. To breakthrough this constraint, we study the task of simultaneous lip reading and devise SimulLR, a simultaneous lip Reading transducer with attention-guided adaptive memory from three aspects: (1) To address the challenge of monotonic alignments while considering the syntactic structure of the generated sentences under simultaneous setting, we build a transducer-based model and design several effective training strategies including CTC pre-training, model warm-up and curriculum learning to promote the training of the lip reading transducer. (2) To learn better spatio-temporal representations for simultaneous encoder, we construct a truncated 3D convolution and time-restricted self-attention layer to perform the frame-to-frame interaction within a video segment containing fixed number of frames. (3) The history information is always limited due to the storage in real-time scenarios, especially for massive video data. Therefore, we devise a novel attention-guided adaptive memory to organize semantic information of history segments and enhance the visual representations with acceptable computation-aware latency. The experiments show that the SimulLR achieves the translation speedup 9.10x compared with the state-of-the-art non-simultaneous methods, and also obtains competitive results, which indicates the effectiveness of our proposed methods.

For face presentation attack detection (PAD), most of the spoofing cues are subtle, local image patterns (e.g., local image distortion, 3D mask edge and cut photo edges). The representations of existing PAD works with simple global pooling method, however, lose the local feature discriminability. In this paper, the VLAD aggregation method is adopted to quantize local features with visual vocabulary locally partitioning the feature space, and hence preserve the local discriminability. We further propose the vocabulary separation and adaptation method to modify VLAD for cross-domain PAD task. The proposed vocabulary separation method divides vocabulary into domain-shared and domain-specific visual words to cope with the diversity of live and attack faces under the cross-domain scenario.The proposed vocabulary adaptation method imitates the maximization step of the k-means algorithm in the end-to-end training, which guarantees the visual words be close to the center of assigned local features and thus brings robust similarity measurement. We give illustrations and extensive experiments to demonstrate the effectiveness of VLAD with the proposed vocabulary separation and adaptation method on standard cross-domain PAD benchmarks. The codes are available at https://github.com/Liubinggunzu/VLAD-VSA.

There is a soaring interest in the news recommendation research scenario due to the information overload. To accurately capture users' interests, we propose to model multi-modal features, in addition to the news titles that are widely used in existing works, for news recommendation. Besides, existing research pays little attention to the click decision-making process in designing multi-modal modeling modules. In this work, inspired by the fact that users make their click decisions mostly based on the visual impression they perceive when browsing news, we propose to capture such visual impression information with visual-semantic modeling for news recommendation. In this paper, we refer to visual impression as the region of the news displayed on the user interface of a news application, which delivers both content and layout information to users. Specifically, we devise the local impression modeling module to simultaneously attend to decomposed details in the impression when understanding the semantic meaning of news title, which could explicitly get close to the process of users reading news. In addition, we inspect the impression from a global view and take structural information, such as the arrangement of different fields and spatial position of different words on the impression, into the modeling of multiple modalities. To accommodate the research of visual impression-aware news recommendation, we extend the text-dominated news recommendation dataset MIND by adding snapshot impression images and will release it to nourish the research field. Extensive comparisons with the state-of-the-art news recommenders along with the in-depth analyses demonstrate the effectiveness of the proposed method and the promising capability of modeling visual impressions for the content-based recommenders.

Sign language translation as a kind of technology with profound social significance has attracted growing researchers' interest in recent years. However, the existing sign language translation methods need to read all the videos before starting the translation, which leads to a high inference latency and also limits their application in real-life scenarios. To solve this problem, we propose SimulSLT, the first end-to-end simultaneous sign language translation model, which can translate sign language videos into target text concurrently. SimulSLT is composed of a text decoder, a boundary predictor, and a masked encoder. We 1) use the wait-k strategy for simultaneous translation. 2) design a novel boundary predictor based on the integrate-and-fire module to output the gloss boundary, which is used to model the correspondence between the sign language video and the gloss. 3) propose an innovative re-encode method to help the model obtain more abundant contextual information, which allows the existing video features to interact fully. The experimental results conducted on the RWTH-PHOENIX-Weather 2014T dataset show that SimulSLT achieves BLEU scores that exceed the latest end-to-end non-simultaneous sign language translation model while maintaining low latency, which proves the effectiveness of our method.

The majority of existing multimodal sequential learning methods focus on how to obtain effective representations and ignore the importance of multimodal fusion. Bilinear attention network (BAN) is a commonly used fusion method, which leverages tensor operations to associate the features of different modalities. However, BAN has a poor compatibility for more modalities, since the computational complexity of the attention map increases exponentially with the number of modalities. Based on this concern, we propose a new method called generalizable multi-linear attention network (MAN), which can associate as many modalities as possible in linear complexity with hierarchical approximation decomposition (HAD). Besides, considering the fact that softmax attention kernels cannot be decomposed as linear operation directly, we adopt the addition random features (ARF) mechanism to approximate the non-linear softmax functions with enough theoretical analysis. We conduct extensive experiments on four datasets of three tasks (multimodal sentiment analysis, multimodal speaker traits recognition, and video retrieval), the experimental results show that MAN could achieve competitive results compared with the state-of-the-art methods, showcasing the effectiveness of the approximation decomposition and addition random features mechanism.

Non-autoregressive text-to-speech (NAR-TTS) models such as FastSpeech 2 and Glow-TTS can synthesize high-quality speech from the given text in parallel. After analyzing two kinds of generative NAR-TTS models (VAE and normalizing flow), we find that: VAE is good at capturing the long-range semantics features (e.g., prosody) even with small model size but suffers from blurry and unnatural results; and normalizing flow is good at reconstructing the frequency bin-wise details but performs poorly when the number of model parameters is limited. Inspired by these observations, to generate diverse speech with natural details and rich prosody using a lightweight architecture, we propose PortaSpeech, a portable and high-quality generative text-to-speech model. Specifically, 1) to model both the prosody and mel-spectrogram details accurately, we adopt a lightweight VAE with an enhanced prior followed by a flow-based post-net with strong conditional inputs as the main architecture. 2) To further compress the model size and memory footprint, we introduce the grouped parameter sharing mechanism to the affine coupling layers in the post-net. 3) To improve the expressiveness of synthesized speech and reduce the dependency on accurate fine-grained alignment between text and speech, we propose a linguistic encoder with mixture alignment combining hard word-level alignment and soft phoneme-level alignment, which explicitly extracts word-level semantic information. Experimental results show that PortaSpeech outperforms other TTS models in both voice quality and prosody modeling in terms of subjective and objective evaluation metrics, and shows only a slight performance degradation when reducing the model parameters to 6.7M (about 4x model size and 3x runtime memory compression ratio compared with FastSpeech 2). Our extensive ablation studies demonstrate that each design in PortaSpeech is effective.

Due to the popularity of video contents on the Internet, the information retrieval between videos and texts has attracted broad interest from researchers, which is a challenging cross-modal retrieval task. A common solution is to learn a joint embedding space to measure the cross-modal similarity. However, many existing approaches either pay more attention to textual information, video information, or cross-modal matching methods, but less to all three. We believe that a good video-text retrieval system should take into account all three points, fully exploiting the semantic information of both modalities and considering a comprehensive match. In this paper, we propose a Hierarchical Cross-Modal Graph Consistency Learning Network (HCGC) for video-text retrieval task, which considers multi-level graph consistency for video-text matching. Specifically, we first construct a hierarchical graph representation for the video, which includes three levels from global to local: video, clips and objects. Similarly, the corresponding text graph is constructed according to the semantic relationships among sentence, actions and entities. Then, in order to learn a better match between the video and text graph, we design three types of graph consistency (both direct and indirect): inter-graph parallel consistency, inter-graph cross consistency and intra-graph cross consistency. Extensive experimental results on different video-text datasets demonstrate the effectiveness of our approach on both text-to-video and video-to-text retrieval.

Learning user representations based on historical behaviors lies at the core of modern recommender systems. Recent advances in sequential recommenders have convincingly demonstrated high capability in extracting effective user representations from the given behavior sequences. Despite significant progress, we argue that solely modeling the observational behaviors sequences may end up with a brittle and unstable system due to the noisy and sparse nature of user interactions logged. In this paper, we propose to learn accurate and robust user representations, which are required to be less sensitive to (attack on) noisy behaviors and trust more on the indispensable ones, by modeling counterfactual data distribution. Specifically, given an observed behavior sequence, the proposed CauseRec framework identifies dispensable and indispensable concepts at both the fine-grained item level and the abstract interest level. CauseRec conditionally samples user concept sequences from the counterfactual data distributions by replacing dispensable and indispensable concepts within the original concept sequence. With user representations obtained from the synthesized user sequences, CauseRec performs contrastive user representation learning by contrasting the counterfactual with the observational. We conduct extensive experiments on real-world public recommendation benchmarks and justify the effectiveness of CauseRec with multi-aspects model analysis. The results demonstrate that the proposed CauseRec outperforms state-of-the-art sequential recommenders by learning accurate and robust user representations.

In recommender systems, modeling user-item behaviors is essential for user representation learning. Existing sequential recommenders consider the sequential correlations between historically interacted items for capturing users’ historical preferences. However, since users’ preferences are by nature time-evolving and diversified, solely modeling the historical preference (without being aware of the time-evolving trends of preferences) can be inferior for recommending complementary or fresh items and thus hurt the effectiveness of recommender systems. In this paper, we bridge the gap between the past preference and potential future preference by proposing the future-aware diverse trends (FAT) framework. By future-aware, for each inspected user, we construct the future sequences from other similar users, which comprise of behaviors that happen after the last behavior of the inspected user, based on a proposed neighbor behavior extractor. By diverse trends, supposing the future preferences can be diversified, we propose the diverse trends extractor and the time-aware mechanism to represent the possible trends of preferences for a given user with multiple vectors. We leverage both the representations of historical preference and possible future trends to obtain the final recommendation. The quantitative and qualitative results from relatively extensive experiments on real-world datasets demonstrate the proposed framework not only outperforms the state-of-the-art sequential recommendation methods across various metrics, but also makes complementary and fresh recommendations.

Video question generation is a challenging task in visual information retrieval, which generates questions given a sequence of video frames. The existing methods mainly tackle the problem of single-turn video question generation, but single-turn conversation usually can't meet the needs of video information acquisition. In this paper, we propose a new framework for single-turn VQG, which introduces attention mechanism to process inference of dialog history. And we introduce selection mechanism to choose from the candidate questions generated by each round of dialog history. In the framework, we leverage a recent video question answering model to predict the answer to the generated question and adopt the answer quality as rewards to fine-tune our model based on a reinforced learning mechanism. We also introduce a new task of multi-turn video question generation (M-VQG), which is generating multiple questions based on dialog history and video information to build conversation step by step. Our method achieves the state-of-the-art performance of the single-turn VQG task on two large-scale datasets, YouTube-Clips and TACoS-MultiLevel, and provides a baseline approach for M-VQG task.

Visual narrating focuses on generating semantic descriptions to summarize visual content of images or videos, e.g., visual captioning and visual storytelling. The challenge mainly lies in how to design a decoder to generate accurate descriptions matching visual content. Recent advances often employ a recurrent neural network (RNN), e.g., Long-Short Term Memory (LSTM), as the decoder. However, RNN is prone to diluting long-term information, which weakens its performance of capturing long-term dependencies. Recent work has demonstrated memory network (MemNet) owns the advantage of storing long-term information. However, as the decoder, it has not been well exploited for visual narrating. The reason partially comes from the difficulty of multi-modal sequential decoding with MemNet. In this article, we devise a novel memory decoder for visual narrating. Concretely, to obtain a better multi-modal representation, we first design a new multi-modal fusion method to fully merge visual and lexical information. Then, based on the fusion result, during decoding, we construct a MemNet-based decoder consisting of multiple memory layers. Particularly, in each layer, we employ a memory set to store previous decoding information and utilize an attention mechanism to adaptively select the information related to the current output. Meanwhile, we also employ a memory set to store the decoding output of each memory layer at the current time step and still utilize an attention mechanism to select the related information. Thus, this decoder alleviates dilution of long-term information. Meanwhile, the hierarchical architecture leverages the latent information of each layer, which is helpful for generating accurate descriptions. Experimental results on two tasks of visual narrating, i.e., video captioning and visual storytelling, show that our decoder could obtain superior results and outperform the performance of conventional RNN-based decoder.

Video question answering is an important task combining both Natural Language Processing and Computer Vision, which requires a machine to obtain a thorough understanding of the video. Most existing approaches simply capture spatio-temporal information in videos by using a combination of recurrent and convolutional neural networks. Nonetheless, most previous work focus on only salient frames or regions, which normally lacks some significant details, such as potential location and action relations. In this paper, we propose a new method called Graph-based Multi-interaction Network for video question answering. In our model, a new attention mechanism named multi-interaction is designed to capture both element-wise and segment-wise sequence interactions simultaneously, which can be found between and inside the multi-modal inputs. Moreover, we propose a graph-based relation-aware neural network to explore a more fine-grained visual representation, which could explore the relationships and dependencies between objects spatially and temporally. We evaluate our method on TGIF-QA and other two video QA datasets. The qualitative and quantitative experimental results show the effectiveness of our model, which achieves state-of-the-art performance.

Vision-language research has become very popular, which focuses on understanding of visual contents, language semantics and relationships between them. Video question answering (Video QA) is one of the typical tasks. Recently, several BERT style pre-training methods have been proposed and shown effectiveness on various vision-language tasks. In this work, we leverage the successful vision-language transformer structure to solve the Video QA problem. However, we do not pre-train it with any video data, because video pre-training requires massive computing resources and is hard to perform with only a few GPUs. Instead, our work aims to leverage image-language pre-training to help with video-language modeling, by sharing a common module design. We further introduce an adaptive spatio-temporal graph to enhance the vision-language representation learning. That is, we adaptively refine the spatio-temporal tubes of salient objects according to their spatio-temporal relations learned through a hierarchical graph convolution process. Finally, we can obtain a number of fine-grained tube-level video object representations, as the visual inputs of the vision-language transformer module. Experiments on three widely used Video QA datasets show that our model achieves the new state-of-the-art results.

Given a natural language description, temporal textual localization aims to localize the most relevant segment in an untrimmed video, which is a natural and imperative extension of temporal action localization. Most existing temporal textual localization works neglect the long-range semantic modeling in video contents and lack accurate textual understanding. Moreover, they remain in single-task learning and fail to exploit multi-view supervised information. Based on these observations, we introduce a novel adversarial bi-directional interaction network, which is a global framework to retrieve the target segment directly. Specifically, we propose a bi-directional attention mechanism to build bi-directional information interaction, which captures long-range semantic dependencies from video context and enhances textual representation learning. After localization, we further advise an auxiliary discriminator network to verify the localization result and boost the performance by adversarial training process. We adopt multi-task learning approach to train our model, including: (1) predicting coordinate probability distribution task, which selects start and end frame to localize target segment; (2) predicting frame-level correlation distribution task, which calculates the correlation between frame and description; (3) auxiliary adversarial learning task, which calculates matched score between localization and description to boost the performance. The extensive experiments on ActivityNet Captions and TACoS show the significant effectiveness and efficiency of our method.

Developing an abstractive text summarization (ATS) system that is capable of generating concise, appropriate, and plausible summaries for the source documents is a long-term goal of artificial intelligence (AI). Recent advances in ATS are overwhelmingly contributed by deep learning techniques, which have taken the state-of-the-art of ATS to a new level. Despite the significant success of previous methods, generating high-quality and human-like abstractive summaries remains a challenge in practice. The human reading cognition, which is essential for reading comprehension and logical thinking, is still relatively new territory and underexplored in deep neural networks. In this article, we propose a novel Hierarchical Human-like deep neural network for ATS (HH-ATS), inspired by the process of how humans comprehend an article and write the corresponding summary. Specifically, HH-ATS is composed of three primary components (i.e., a knowledge-aware hierarchical attention module, a multitask learning module, and a dual discriminator generative adversarial network), which mimic the three stages of human reading cognition (i.e., rough reading, active reading, and postediting). Experimental results on two benchmark data sets (CNN/Daily Mail and Gigaword) demonstrate that HH-ATS consistently and substantially outperforms the compared methods.

Real-time event summarization (RES) aims at extracting a handful of document updates from an overwhelming document stream as the real-time event summary that tracks and summarizes the evolving event of interest. It has been attracting much attention, especially with the growth of streaming applications. Despite the effectiveness of previous studies, obtaining relevant, nonredundant, and timely event summaries remains challenging in real-life applications. This study proposes an effective Hybrid learning model for RES (HRES), which attempts to resolve all three challenges (i.e., nonredundancy, relevance, and timeliness) of RES in a unified framework. The main idea is to: 1) exploit the factual background knowledge from the knowledge base (KB) to capture the informative knowledge and implicit information from the input document/query for better text matching; 2) design a memory network to memorize the input facts temporally from the historical document stream and avoid pushing redundant facts in subsequent timesteps; 3) leverage relevance prediction as an auxiliary task to strengthen the document modeling and help to extract relevant documents; and 4) consider both historical dependencies and future uncertainty of the real-time document stream by exploiting the reinforcement learning technique. Extensive experiments demonstrate that HRES has robust superiority over competitors and gains the state-of-the-art results.

Video moment retrieval is to search the moment that is most relevant to the given natural language query. Existing methods are mostly trained in a fully-supervised setting, which requires the full annotations of temporal boundary for each query. However, manually labeling the annotations is actually time-consuming and expensive. In this paper, we propose a novel weakly-supervised moment retrieval framework requiring only coarse video-level annotations for training. Specifically, we devise a proposal generation module that aggregates the context information to generate and score all candidate proposals in one single pass. We then devise an algorithm that considers both exploitation and exploration to select top-K proposals. Next, we build a semantic completion module to measure the semantic similarity between the selected proposals and query, compute reward and provide feedbacks to the proposal generation module for scoring refinement. Experiments on the ActivityCaptions and Charades-STA demonstrate the effectiveness of our proposed method.

Image captioning is usually built on either generation-based or retrieval-based approaches. Both ways have certain strengths but suffer from their own limitations. In this paper, we propose an Interactive Dual Generative Adversarial Network (IDGAN) for image captioning, which mutually combines the retrieval-based and generation-based methods to learn a better image captioning ensemble. IDGAN consists of two generators and two discriminators, where the generation- and retrieval-based generators mutually benefit from each other's complementary targets that are learned from two dual adversarial discriminators. Specifically, the generation- and retrieval-based generators provide improved synthetic and retrieved candidate captions with informative feedback signals from the two respective discriminators that are trained to distinguish the generated captions from the true captions and assign top rankings to true captions respectively, thus featuring the merits of both retrieval-based and generation-based approaches. Extensive experiments on MSCOCO dataset demonstrate that the proposed IDGAN model significantly outperforms the compared methods for image captioning.

Previous approaches for video summarization mainly concentrate on finding the most diverse and representative visual contents as video summary without considering the user's preference. This paper addresses the task of query-focused video summarization, which takes user's query and a long video as inputs and aims to generate a query-focused video summary. In this paper, we consider the task as a problem of computing similarity between video shots and query. To this end, we propose a method, named Convolutional Hierarchical Attention Network (CHAN), which consists of two parts: feature encoding network and query-relevance computing module. In the encoding network, we employ a convolutional network with local self-attention mechanism and query-aware global attention mechanism to learns visual information of each shot. The encoded features will be sent to query-relevance computing module to generate query-focused video summary. Extensive experiments on the benchmark dataset demonstrate the competitive performance and show the effectiveness of our approach.

Real-time event summarization is an essential task in natural language processing and information retrieval areas. Despite the progress of previous work, generating relevant, non-redundant, and timely event summaries remains challenging in practice. In this paper, we propose a Deep Reinforcement learning framework for real-time Event Summarization (DRES), which shows promising performance for resolving all three challenges (i.e., relevance, non-redundancy, timeliness) in a unified framework. Specifically, we (i) devise a hierarchical cross-attention network with intra- and inter-document attentions to integrate important semantic features within and between the query and input document for better text matching. In addition, relevance prediction is leveraged as an auxiliary task to strengthen the document modeling and help to extract relevant documents; (ii) propose a multi-topic dynamic memory network to capture the sequential patterns of different topics belonging to the event of interest and temporally memorize the input facts from the evolving document stream, avoiding extracting redundant information at each time step; (iii) consider both historical dependencies and future uncertainty of the document stream for generating relevant and timely summaries by exploiting the reinforcement learning technique. Experimental results on two real-world datasets have demonstrated the advantages of DRES model with significant improvement in generating relevant, non-redundant, and timely event summaries against the state-of-the-arts.

In this work, we develop SimulSpeech, an endto-end simultaneous speech to text translation system which translates speech in source language to text in target language concurrently.SimulSpeech consists of a speech encoder, a speech segmenter and a text decoder, where 1) the segmenter builds upon the encoder and leverages a connectionist temporal classification (CTC) loss to split the input streaming speech in real time, 2) the encoder-decoder attention adopts a wait-k strategy for simultaneous translation.SimulSpeech is more challenging than previous cascaded systems (with simultaneous automatic speech recognition (ASR) and simultaneous neural machine translation (NMT)).We introduce two novel knowledge distillation methods to ensure the performance: 1) Attention-level knowledge distillation transfers the knowledge from the multiplication of the attention matrices of simultaneous NMT and ASR models to help the training of the attention mechanism in SimulSpeech; 2) Data-level knowledge distillation transfers the knowledge from the full-sentence NMT model and also reduces the complexity of data distribution to help on the optimization of Simul-Speech.Experiments on MuST-C English-Spanish and English-German spoken language translation datasets show that SimulSpeech achieves reasonable BLEU scores and lower delay compared to full-sentence end-to-end speech to text translation (without simultaneous translation), and better performance than the two-stage cascaded simultaneous translation model in terms of BLEU scores and translation delay.

Non-autoregressive (NAR) models generate all the tokens of a sequence in parallel, resulting in faster generation speed compared to their autoregressive (AR) counterparts but at the cost of lower accuracy.Different techniques including knowledge distillation and source-target alignment have been proposed to bridge the gap between AR and NAR models in various tasks such as neural machine translation (NMT), automatic speech recognition (ASR), and text to speech (TTS).With the help of those techniques, NAR models can catch up with the accuracy of AR models in some tasks but not in some others.In this work, we conduct a study to understand the difficulty of NAR sequence generation and try to answer: (1) Why NAR models can catch up with AR models in some tasks but not all?(2) Why techniques like knowledge distillation and source-target alignment can help NAR models.Since the main difference between AR and NAR models is that NAR models do not use dependency among target tokens while AR models do, intuitively the difficulty of NAR sequence generation heavily depends on the strongness of dependency among target tokens.To quantify such dependency, we propose an analysis model called CoMMA to characterize the difficulty of different NAR sequence generation tasks.We have several interesting findings: 1) Among the NMT, ASR and TTS tasks, ASR has the most target-token dependency while TTS has the least.2) Knowledge distillation reduces the target-token dependency in target sequence and thus improves the accuracy of NAR models.3) Source-target alignment constraint encourages dependency * Equal contribution.

In this paper, we consider a novel task, Spatio-Temporal Video Grounding for Multi-Form Sentences (STVG). Given an untrimmed video and a declarative/interrogative sentence depicting an object, STVG aims to localize the spatio-temporal tube of the queried object. STVG has two challenging settings: (1) We need to localize spatio-temporal object tubes from untrimmed videos, where the object may only exist in a very small segment of the video; (2) We deal with multi-form sentences, including the declarative sentences with explicit objects and interrogative sentences with unknown objects. Existing methods cannot tackle the STVG task due to the ineffective tube pre-generation and the lack of object relationship modeling. Thus, we then propose a novel Spatio-Temporal Graph Reasoning Network (STGRN) for this task. First, we build a spatio-temporal region graph to capture the region relationships with temporal object dynamics, which involves the implicit and explicit spatial subgraphs in each frame and the temporal dynamic subgraph across frames. We then incorporate textual clues into the graph and develop the multi-step cross-modal graph reasoning. Next, we introduce a spatio-temporal localizer with a dynamic selection method to directly retrieve the spatio-temporal tubes without tube pre-generation. Moreover, we contribute a large-scale video grounding dataset VidSTG based on video relation dataset VidOR. The extensive experiments demonstrate the effectiveness of our method.

Non-autoregressive translation (NAT) achieves faster inference speed but at the cost of worse accuracy compared with autoregressive translation (AT). Since AT and NAT can share model structure and AT is an easier task than NAT due to the explicit dependency on previous target-side tokens, a natural idea is to gradually shift the model training from the easier AT task to the harder NAT task. To smooth the shift from AT training to NAT training, in this paper, we introduce semi-autoregressive translation (SAT) as intermediate tasks. SAT contains a hyperparameter k, and each k value defines a SAT task with different degrees of parallelism. Specially, SAT covers AT and NAT as its special cases: it reduces to AT when k=1 and to NAT when k=N (N is the length of target sentence). We design curriculum schedules to gradually shift k from 1 to N, with different pacing functions and number of tasks trained at the same time. We called our method as task-level curriculum learning for NAT (TCL-NAT). Experiments on IWSLT14 De-En, IWSLT16 En-De, WMT14 En-De and De-En datasets show that TCL-NAT achieves significant accuracy improvements over previous NAT baselines and reduces the performance gap between NAT and AT models to 1-2 BLEU points, demonstrating the effectiveness of our proposed method.

Spatio-temporal video grounding aims to retrieve the spatio-temporal tube of a queried object according to the given sentence. Currently, most existing grounding methods are restricted to well-aligned segment-sentence pairs. In this paper, we explore spatio-temporal video grounding on unaligned data and multi-form sentences. This challenging task requires to capture critical object relations to identify the queried target. However, existing approaches cannot distinguish notable objects and remain in ineffective relation modeling between unnecessary objects. Thus, we propose a novel object-aware multi-branch relation network for object-aware relation discovery. Concretely, we first devise multiple branches to develop object-aware region modeling, where each branch focuses on a crucial object mentioned in the sentence. We then propose multi-branch relation reasoning to capture critical object relationships between the main branch and auxiliary branches. Moreover, we apply a diversity loss to make each branch only pay attention to its corresponding object and boost multi-branch learning. The extensive experiments show the effectiveness of our proposed method.

In this paper, we develop DeepSinger, a multi-lingual multi-singer singing voice synthesis (SVS) system, which is built from scratch using singing training data mined from music websites. The pipeline of DeepSinger consists of several steps, including data crawling, singing and accompaniment separation, lyrics-to-singing alignment, data filtration, and singing modeling. Specifically, we design a lyrics-to-singing alignment model to automatically extract the duration of each phoneme in lyrics starting from coarse-grained sentence level to fine-grained phoneme level, and further design a multi-lingual multi-singer singing model based on a feed-forward Transformer to directly generate linear-spectrograms from lyrics, and synthesize voices using Griffn-Lim. DeepSinger has several advantages over previous SVS systems: 1) to the best of our knowledge, it is the first SVS system that directly mines training data from music websites, 2) the lyrics-to-singing alignment model further avoids any human efforts for alignment labeling and greatly reduces labeling cost, 3) the singing model based on a feed-forward Transformer is simple and efficient, by removing the complicated acoustic feature modeling in parametric synthesis and leveraging a reference encoder to capture the timbre of a singer from noisy singing data, and 4) it can synthesize singing voices in multiple languages and multiple singers. We evaluate DeepSinger on our mined singing dataset that consists of about 92 hours data from 89 singers on three languages (Chinese, Cantonese and English). The results demonstrate that with the singing data purely mined from the Web, DeepSinger can synthesize high-quality singing voices in terms of both pitch accuracy and voice naturalness. Our audio samples are shown in https://speechresearch.github.io/deepsinger/.

In e-commerce, consumer-generated videos, which in general deliver consumers' individual preferences for the different aspects of certain products, are massive in volume. To recommend these videos to potential consumers more effectively, diverse and catchy video titles are critical. However, consumer-generated videos seldom accompany appropriate titles. To bridge this gap, we integrate comprehensive sources of information, including the content of consumer-generated videos, the narrative comment sentences supplied by consumers, and the product attributes, in an end-to-end modeling framework. Although automatic video titling is very useful and demanding, it is much less addressed than video captioning. The latter focuses on generating sentences that describe videos as a whole while our task requires the product-aware multi-grained video analysis. To tackle this issue, the proposed method consists of two processes, i.e., granular-level interaction modeling and abstraction-level story-line summarization. Specifically, the granular-level interaction modeling first utilizes temporal-spatial landmark cues, descriptive words, and abstractive attributes to builds three individual graphs and recognizes the intra-actions in each graph through Graph Neural Networks (GNN). Then the global-local aggregation module is proposed to model inter-actions across graphs and aggregate heterogeneous graphs into a holistic graph representation. The abstraction-level story-line summarization further considers both frame-level video features and the holistic graph to utilize the interactions between products and backgrounds, and generate the story-line topic of the video. We collect a large-scale dataset accordingly from real-world data in Taobao, a world-leading e-commerce platform, and will make the desensitized version publicly available to nourish further development of the research community. Relatively extensive experiments on various datasets demonstrate the efficacy of the proposed method.

Given a partially masked image, image inpainting aims to complete the missing region and output a plausible image. Most existing image inpainting methods complete the missing region by expanding or borrowing information from the surrounding source region, which work well when the original content in the missing region is similar to the surrounding source region. Unsatisfactory results will be generated if there is no sufficient contextual information can be referenced from source region. Besides, the inpainting results should be diverse and this kind of diversity should be controllable. Based on these observations, we propose a new inpainting problem that introduces text as a kind of guidance to direct and control the inpainting process. The main difference between this problem and previous works is that we need ensure the result to be consistent with not only the source region but also the textual guidance during inpainting. By this way, we want to avoid the unreasonable completion and meanwhile make it controllable. We propose a progressively coarse-to-fine cross-modal generative network and adopt the text-image-text training schema to generate visually consistent and semantically coherent images. Extensive quantitative and qualitative experiments on two public datasets with captions demonstrate the effectiveness of our method.

Lipreading is an impressive technique and there has been a definite improvement of accuracy in recent years. However, existing methods for lipreading mainly build on autoregressive (AR) model, which generate target tokens one by one and suffer from high inference latency. To breakthrough this constraint, we propose FastLR, a non-autoregressive (NAR) lipreading model which generates all target tokens simultaneously. NAR lipreading is a challenging task that has many difficulties: 1) the discrepancy of sequence lengths between source and target makes it difficult to estimate the length of the output sequence; 2) the conditionally independent behavior of NAR generation lacks the correlation across time which leads to a poor approximation of target distribution; 3) the feature representation ability of encoder can be weak due to lack of effective alignment mechanism; and 4) the removal of AR language model exacerbates the inherent ambiguity problem of lipreading. Thus, in this paper, we introduce three methods to reduce the gap between FastLR and AR model: 1) to address challenges 1 and 2, we leverage integrate-and-fire (I&F) module to model the correspondence between source video frames and output text sequence. 2) To tackle challenge 3, we add an auxiliary connectionist temporal classification (CTC) decoder to the top of the encoder and optimize it with extra CTC loss. We also add an auxiliary autoregressive decoder to help the feature extraction of encoder. 3) To overcome challenge 4, we propose a novel Noisy Parallel Decoding (NPD) for I&F and bring Byte-Pair Encoding (BPE) into lipreading. Our experiments exhibit that FastLR achieves the speedup up to 10.97× comparing with state-of-the-art lipreading model with slight WER absolute increase of 1.5% and 5.5% on GRID and LRS2 lipreading datasets respectively, which demonstrates the effectiveness of our proposed method.

In pop music, accompaniments are usually played by multiple instruments (tracks) such as drum, bass, string and guitar, and can make a song more expressive and contagious by arranging together with its melody. Previous works usually generate multiple tracks separately and the music notes from different tracks not explicitly depend on each other, which hurts the harmony modeling. To improve harmony, in this paper, we propose a novel MUlti-track MIDI representation (MuMIDI), which enables simultaneous multi-track generation in a single sequence and explicitly models the dependency of the notes from different tracks. While this greatly improves harmony, unfortunately, it enlarges the sequence length and brings the new challenge of long-term music modeling. We further introduce two new techniques to address this challenge: 1) We model multiple note attributes (e.g., pitch, duration, velocity) of a musical note in one step instead of multiple steps, which can shorten the length of a MuMIDI sequence. 2) We introduce extra long-context as memory to capture long-term dependency in music. We call our system for pop music accompaniment generation as PopMAG. We evaluate PopMAG on multiple datasets (LMD, FreeMidi and CPMD, a private dataset of Chinese pop songs) with both subjective and objective metrics. The results demonstrate the effectiveness of PopMAG for multi-track harmony modeling and long-term context modeling. Specifically, PopMAG wins 42%/38%/40% votes when comparing with ground truth musical pieces on LMD, FreeMidi and CPMD datasets respectively and largely outperforms other state-of-the-art music accompaniment generation models and multi-track MIDI representations in terms of subjective and objective metrics.

In this paper, we propose to investigate the problem of out-of-domain visio-linguistic pretraining, where the pretraining data distribution differs from that of downstream data on which the pretrained model will be fine-tuned. Existing methods for this problem are purely likelihood-based, leading to the spurious correlations and hurt the generalization ability when transferred to out-of-domain downstream tasks. By spurious correlation, we mean that the conditional probability of one token (object or word) given another one can be high (due to the dataset biases) without robust (causal) relationships between them. To mitigate such dataset biases, we propose a Deconfounded Visio-Linguistic Bert framework, abbreviated as DeVLBert, to perform intervention-based learning. We borrow the idea of the backdoor adjustment from the research field of causality and propose several neural-network based architectures for Bert-style out-of-domain pretraining. The quantitative results on three downstream tasks, Image Retrieval (IR), Zero-shot IR, and Visual Question Answering, show the effectiveness of DeVLBert by boosting generalization ability.

Video moment retrieval aims to localize the target moment in an video according to the given sentence. The weak-supervised setting only provides the video-level sentence annotations during training. Most existing weak-supervised methods apply a MIL-based framework to develop inter-sample confrontment, but ignore the intra-sample confrontment between moments with semantically similar contents. Thus, these methods fail to distinguish the target moment from plausible negative moments. In this paper, we propose a novel Regularized Two-Branch Proposal Network to simultaneously consider the inter-sample and intra-sample confrontments. Concretely, we first devise a language-aware filter to generate an enhanced video stream and a suppressed video stream. We then design the sharable two-branch proposal module to generate positive proposals from the enhanced stream and plausible negative proposals from the suppressed one for sufficient confrontment. Further, we apply the proposal regularization to stabilize the training process and improve model performance. The extensive experiments show the effectiveness of our method. Our code is released at here.

In e-commerce, a growing number of user-generated videos are used for product promotion. How to generate video descriptions that narrate the user-preferred product characteristics depicted in the video is vital for successful promoting. Traditional video captioning methods, which focus on routinely describing what exists and happens in a video, are not amenable for product-oriented video captioning. To address this problem, we propose a product-oriented video captioner framework, abbreviated as Poet. Poet firstly represents the videos as product-oriented spatial-temporal graphs. Then, based on the aspects of the video-associated product, we perform knowledge-enhanced spatial-temporal inference on those graphs for capturing the dynamic change of fine-grained product-part characteristics. The knowledge leveraging module in Poet differs from the traditional design by performing knowledge filtering and dynamic memory modeling. We show that Poet achieves consistent performance improvement over previous methods concerning generation quality, product aspects capturing, and lexical diversity. Experiments are performed on two product-oriented video captioning datasets, buyer-generated fashion video dataset (BFVD) and fan-generated fashion video dataset (FFVD), collected from Mobile Taobao. We will release the desensitized datasets to promote further investigations on both video captioning and general video analysis problems.

Weakly-supervised vision-language grounding aims to localize a target moment in a video or a specific region in an image according to the given sentence query, where only video-level or image-level sentence annotations are provided during training. Most existing approaches employ the MIL-based or reconstruction-based paradigms for the WSVLG task, but the former heavily depends on the quality of randomly-selected negative samples and the latter cannot directly optimize the visual-textual alignment score. In this paper, we propose a novel Counterfactual Contrastive Learning (CCL) to develop sufficient contrastive training between counterfactual positive and negative results, which are based on robust and destructive counterfactual transformations. Concretely, we design three counterfactual transformation strategies from the feature-, interaction- and relation-level, where the feature-level method damages the visual features of selected proposals, interaction-level approach confuses the vision-language interaction and relation-level strategy destroys the context clues in proposal relationships. Extensive experiments on five vision-language grounding datasets verify the effectiveness of our CCL paradigm.

Sequential recommender systems (SRS) have become the key technology in capturing user's dynamic interests and generating high-quality recommendations. Current state-of-the-art sequential recommender models are typically based on a sandwich-structured deep neural network, where one or more middle (hidden) layers are placed between the input embedding layer and output softmax layer. In general, these models require a large number of parameters to obtain optimal performance. Despite the effectiveness, at some point, further increasing model size may be harder for model deployment in resource-constraint devices. To resolve the issues, we propose a compressed sequential recommendation framework, termed as CpRec, where two generic model shrinking techniques are employed. Specifically, we first propose a block-wise adaptive decomposition to approximate the input and softmax matrices by exploiting the fact that items in SRS obey a long-tailed distribution. To reduce the parameters of the middle layers, we introduce three layer-wise parameter sharing schemes. We instantiate CpRec using deep convolutional neural network with dilated kernels given consideration to both recommendation accuracy and efficiency. By the extensive ablation studies, we demonstrate that the proposed CpRec can achieve up to 4~8 times compression rates in real-world SRS datasets. Meanwhile, CpRec is faster during training & inference, and in most cases outperforms its uncompressed counterpart.

In the process of visual perception, humans perceive not only the appearance of objects existing in a place but also their relationships (e.g. spatial layout). However, the dominant works on visual place recognition are always based on the assumption that two images depict the same place if they contain enough similar objects, while the relation information is neglected. In this paper, we propose a regional relation module which models the regional relationships and converts the convolutional feature maps to the relational feature maps. We further design a cascaded pooling method to get discriminative relation descriptors by preventing the influence of confusing relations and preserving as much useful information as possible. Extensive experiments on two place recognition benchmarks demonstrate that training with the proposed regional relation module improves the appearance descriptors and the relation descriptors are complementary to appearance descriptors. When these two kinds of descriptors are concatenated together, the resulting combined descriptors outperform the state-of-the-art methods.

Video dialog is a new and challenging task, which requires an AI agent to maintain a meaningful dialog with humans in natural language about video contents. Specifically, given a video, a dialog history and a new question about the video, the agent has to combine video information with dialog history to infer the answer. However, the existing methods of image dialog and video question answering, which fail to process the complexity of video information and establish the logical dependency of history contexts, are inappropriate to be applied directly to video dialog. In this paper, we propose a novel approach for video dialog called multi-grained convolutional self-attention context network, which combines video information with dialog history. Instead of using RNN to encode the sequence information, we design a multi-grained convolutional self-attention mechanism to capture both element and segment level interactions that contain multi-grained sequence information. Moreover, a hierarchical dialog history encoder is designed to learn the context-aware question representation. Finally, we establish two decoders in multiple-choice and open-ended forms respectively, which utilize different strategies to get the multi-model context-aware video representation and to generate human-like answers. We evaluate our method on two large-scale datasets. Due to the flexibility and parallelism of the new attention mechanism, our method can achieve higher time efficiency, and the extensive experiments also show the effectiveness of our method.

Temporal link prediction in dynamic networks has attracted increasing attention recently due to its valuable real-world applications. The primary challenge of temporal link prediction is to capture the spatial-temporal patterns and high nonlinearity of dynamic networks. Inspired by the success of image generation, we convert the dynamic network into a sequence of static images and formulate the temporal link prediction as a conditional image generation problem. We propose a novel deep generative framework, called NetworkGAN, to tackle the challenging temporal link prediction task efficiently, which simultaneously models the spatial and temporal features in the dynamic networks via deep learning techniques. The proposed NetworkGAN inherits the advantages of the graph convolutional network (GCN), the temporal matrix factorization (TMF), the long short-term memory network (LSTM), and the generative adversarial network (GAN). Specifically, an attentive GCN is first designed to automatically learn the spatial features of dynamic networks. Second, we propose a TMF enhanced attentive LSTM (TMF-LSTM) to capture the temporal dependencies and evolutionary patterns of dynamic networks, which predicts the network snapshot at next timestamp based on the network snapshots observed at previous timestamps. Furthermore, we employ a GAN framework to further refine the performance of temporal link prediction by using a discriminative model to guide the training of the deep generative model (i.e., TMF-LSTM) in an adversarial process. To verify the effectiveness of the proposed model, we conduct extensive experiments on five real-world datasets. Experimental results demonstrate the significant advantages of NetworkGAN compared to other strong competitors.

Movie recommendation systems provide users with ranked lists of movies based on individual's preferences and constraints. Two types of models are commonly used to generate ranking results: 1) long-term models and 2) session-based models. The long-term-based models represent the interactions between users and movies that are supposed to change slowly across time, while the session-based models encode the information of users' interests and changing dynamics of movies' attributes in short terms. In this paper, we propose the LSIC model, leveraging long and short-term information for content-aware movie recommendation using adversarial training. In the adversarial process, we train a generator as an agent of reinforcement learning which recommends the next movie to a user sequentially. We also train a discriminator which attempts to distinguish the generated list of movies from the real records. The poster information of movies is integrated to further improve the performance of movie recommendation, which is specifically essential when few ratings are available. The experiments demonstrate that the proposed model has robust superiority over competitors and achieves the state-of-the-art results.

Moment retrieval aims to localize the most relevant moment in an untrimmed video according to the given natural language query. Existing works often only focus on one aspect of this emerging task, such as the query representation learning, video context modeling or multi-modal fusion, thus fail to develop a comprehensive system for further performance improvement. In this paper, we introduce a novel Cross-Modal Interaction Network (CMIN) to consider multiple crucial factors for this challenging task, including the syntactic dependencies of natural language queries, long-range semantic dependencies in video context and the sufficient cross-modal interaction. Specifically, we devise a syntactic GCN to leverage the syntactic structure of queries for fine-grained representation learning and propose a multi-head self-attention to capture long-range semantic dependencies from video context. Next, we employ a multi-stage cross-modal interaction to explore the potential relations of video and query contents, and we also consider query reconstruction from the cross-modal representations of target moment as an auxiliary task to strengthen the cross-modal representations. The extensive experiments on ActivityNet Captions and TACoS demonstrate the effectiveness of our proposed method.

This paper addresses the task of query-focused video summarization, which takes user queries and long videos as inputs and generates query-focused video summaries. Compared to video summarization, which mainly concentrates on finding the most diverse and representative visual contents as a summary, the task of query-focused video summarization considers the user's intent and the semantic meaning of generated summary. In this paper, we propose a method, named query-biased self-attentive network (QSAN) to tackle this challenge. Our key idea is to utilize the semantic information from video descriptions to generate a generic summary and then to combine the information from the query to generate a query-focused summary. Specifically, we first propose a hierarchical self-attentive network to model the relative relationship at three levels, which are different frames from a segment, different segments of the same video, textual information of video description and its related visual contents. We train the model on video caption dataset and employ a reinforced caption generator to generate a video description, which can help us locate important frames or shots. Then we build a query-aware scoring module to compute the query-relevant score for each shot and generate the query-focused summary. Extensive experiments on the benchmark dataset demonstrate the competitive performance of our approach compared to some methods.

Image captioning, which aims to generate a sentence to describe the key content of a query image, is an important but challenging task. Existing image captioning approaches can be categorised into two types: generation-based methods and retrieval-based methods. Retrieval-based methods describe images by retrieving pre-existing captions from a repository. Generation-based methods synthesize a new sentence that verbalizes the query image. Both ways have certain advantages but suffer from their own disadvantages. In the paper, we propose a novel EnsCaption model, which aims at enhancing an ensemble of retrieval-based and generation-based image captioning methods through a novel dual generator generative adversarial network. Specifically, EnsCaption is composed of a caption generation model that synthesizes tailored captions for the query image, a caption re-ranking model that retrieves the best-matching caption from a candidate caption pool consisting of generated captions and pre-retrieved captions, and a discriminator that learns the multi-level difference between the generated/retrieved captions and the ground-truth captions. During the adversarial training process, the caption generation model and the caption re-ranking model provide improved synthetic and retrieved candidate captions with high ranking scores from the discriminator, while the discriminator based on multi-level ranking is trained to assign low ranking scores to the generated and retrieved image captions. Our model absorbs the merits of both generation-based and retrieval-based approaches. We conduct comprehensive experiments to evaluate the performance of EnsCaption on two benchmark datasets: MSCOCO and Flickr-30K. Experimental results show that EnsCaption achieves impressive performance compared to the strong baseline methods.

As a challenging task in visual information retrieval, open-ended long-form video question answering automatically generates the natural language answer from the referenced video content according to the given question. However, the existing video question answering works mainly focus on the short-form video, which may be ineffectively applied for long-form video question answering directly, due to the insufficiency of modeling the semantic representation of long-form video content. In this paper, we study the problem of open-ended long-form video question answering from the viewpoint of hierarchical multimodal conditional adversarial network learning. We propose the hierarchical attentional encoder network to learn the joint representation of long-form video content and given question with adaptive video segmentation. We then devise the reinforced decoder network to generate the natural language answer for openended video question answering with multi-modal conditional adversarial network learning. We construct three large-scale open-ended video question answering datasets. The extensive experiments validate the effectiveness of our method.

Game bots are automated programs that assist cheating users, leading to an imbalance in the game ecosystem and the collapse of user interest. Online games provide immersive gaming experience and attract many loyal fans. However, game bots have proliferated in volume and method, evolving with the real-world detection methods and showing strong diversity, leaving game bot detection efforts extremely difficult. Existing game bot detection techniques mostly rely on handcrafted features or time-series based features instead of fully utilizing player behavior sequences. In this regard, a more reasonable way should be learning user patterns from player behavior sequences when facing the fast-changing nature of game bots. Here we propose a general game bot detection framework for massively multiplayer online role playing games termed NGUARD+ (denoting NetEase Games’ Guard), which captures user patterns in order to identify game bots from player behavior sequences. NGUARD+ mainly employs attention-based methods to automatically differentiate game bots from humans. We provide a combination of supervised and unsupervised methods for game bot detection to detect game bots and new type of game bots even when the labels of game bots are limited. Specifically, we propose the following two variants for attention-based sequence modeling: Attention based Bidirectional Long Short-Term Memory Networks (ABLSTM) and Hierarchical Self-Attention Network (HSAN) as our supervised models. ABLSTM is keen on inducing certain inductive biases which makes learning more reasonable as well as capturing local dependency and global information, while HSAN could handle much longer behavior sequences with less memory and higher computational efficiency. Experiments conducted on a real-world dataset show that NGUARD+ can achieve remarkable performance improvement compared to traditional methods. Moreover, NGUARD+ can reveal outstanding robustness for game bots in mutated patterns and even in completely unseen patterns.

Video Question Answering (VideoQA) is the extension of image question answering (ImageQA) in the video domain. Methods are required to give the correct answer after analyzing the provided video and question in this task. Comparing to ImageQA, the most distinctive part is the media type. Both tasks require the understanding of visual media, but VideoQA is much more challenging, mainly because of the complexity and diversity of videos. Particularly, working with the video needs to model its inherent temporal structure and analyze the diverse information it contains. In this article, we propose to tackle the task from a multichannel perspective. Appearance, motion, and audio features are extracted from the video, and question-guided attentions are refined to generate the expressive clues that support the correct answer. We also incorporate the relevant text information acquired from Wikipedia as an attempt to extend the capability of the method. Experiments on TGIF-QA and ActivityNet-QA datasets show the advantages of our method compared to existing methods. We also demonstrate the effectiveness and interpretability of our method by analyzing the refined attention weights during the question-answering procedure.

Online Shopping has become a part of our daily routine, but it still cannot offer intuitive experience as store shopping. Nowadays, most e-commerce Websites offer a Question Answering (QA) system that allows users to consult other users who have purchased the product. However, users still need to wait patiently for others’ replies. In this paper, we investigate how to provide a quick response to the asker by plausible answer identification from product reviews. By analyzing the similarity and discrepancy between explicit answers and reviews that can be answers, a novel multi-task deep learning method with carefully designed attention mechanisms is developed. The method can well exploit large amounts of user generated QA data and a few manually labeled review data to address the problem. Experiments on data collected from Amazon demonstrate its effectiveness and superiority over competitive baselines.

End-to-End deep learning approaches for Automatic Speech Recognition (ASR) has been a new trend. In those approaches, starting active in many areas, language model can be considered as an important and effective method for semantic error correction. Many existing systems use one language model. In this paper, however, multiple language models (LMs) are applied into decoding. One LM is used for selecting appropriate answers and others, considering both context and grammar, for further decision. Experiment on a general location-based dataset show the effectiveness of our method.

The recent artificial intelligence studies have witnessed great interest in abstractive text summarization. Although remarkable progress has been made by deep neural network based methods, generating plausible and high-quality abstractive summaries remains a challenging task. The human-like reading strategy is rarely explored in abstractive text summarization, which however is able to improve the effectiveness of the summarization by considering the process of reading comprehension and logical thinking. Motivated by the humanlike reading strategy that follows a hierarchical routine, we propose a novel Hybrid learning model for Abstractive Text Summarization (HATS). The model consists of three major components, a knowledge-based attention network, a multitask encoder-decoder network, and a generative adversarial network, which are consistent with the different stages of the human-like reading strategy. To verify the effectiveness of HATS, we conduct extensive experiments on two real-life datasets, CNN/Daily Mail and Gigaword datasets. The experimental results demonstrate that HATS achieves impressive results on both datasets.

Recent developments in modeling language and vision have been successfully applied to image question answering. It is both crucial and natural to extend this research direction to the video domain for video question answering (VideoQA). Compared to the image domain where large scale and fully annotated benchmark datasets exists, VideoQA datasets are limited to small scale and are automatically generated, etc. These limitations restrict their applicability in practice. Here we introduce ActivityNet-QA, a fully annotated and large scale VideoQA dataset. The dataset consists of 58,000 QA pairs on 5,800 complex web videos derived from the popular ActivityNet dataset. We present a statistical analysis of our ActivityNet-QA dataset and conduct extensive experiments on it by comparing existing VideoQA baselines. Moreover, we explore various video representation strategies to improve VideoQA performance, especially for long videos.

Hierarchical multi-label text classification (HMTC) is a fundamental but challenging task of numerous applications (e.g., patent annotation), where documents are assigned to multiple categories stored in a hierarchical structure. Categories at different levels of a document tend to have dependencies. However, the majority of prior studies for the HMTC task employ classifiers to either deal with all categories simultaneously or decompose the original problem into a set of flat multi-label classification subproblems, ignoring the associations between texts and the hierarchical structure and the dependencies among different levels of the hierarchical structure. To that end, in this paper, we propose a novel framework called Hierarchical Attention-based Recurrent Neural Network (HARNN) for classifying documents into the most relevant categories level by level via integrating texts and the hierarchical category structure. Specifically, we first apply a documentation representing layer for obtaining the representation of texts and the hierarchical structure. Then, we develop an hierarchical attention-based recurrent layer to model the dependencies among different levels of the hierarchical structure in a top-down fashion. Here, a hierarchical attention strategy is proposed to capture the associations between texts and the hierarchical structure. Finally, we design a hybrid method which is capable of predicting the categories of each level while classifying all categories in the entire hierarchical structure precisely. Extensive experimental results on two real-world datasets demonstrate the effectiveness and explanatory power of HARNN.

In this paper, we propose a multi-task learning approach for cross-modal image-text retrieval. First, a correlation network is proposed for relation recognition task, which helps learn the complicated relations and common information of different modalities. Then, we propose a correspondence cross-modal autoencoder for cross-modal input reconstruction task, which helps correlate the hidden representations of two uni-modal autoencoders. In addition, to further improve the performance of cross-modal retrieval, two regularization terms (variance and consistency constraints) are introduced to the cross-modal embeddings such that the learned common information has large variance and is modality invariant. Finally, to enable large-scale cross-modal similarity search, a flexible binary transform network is designed to convert the text and image embeddings into binary codes. Extensive experiments on two benchmark datasets demonstrate that our model has robust superiority over the compared strong baseline methods. Source code is available at \urlhttps://github.com/daerv/DAEVR.

We propose a self-supervised spatiotemporal learning technique which leverages the chronological order of videos. Our method can learn the spatiotemporal representation of the video by predicting the order of shuffled clips from the video. The category of the video is not required, which gives our technique the potential to take advantage of infinite unannotated videos. There exist related works which use frames, while compared to frames, clips are more consistent with the video dynamics. Clips can help to reduce the uncertainty of orders and are more appropriate to learn a video representation. The 3D convolutional neural networks are utilized to extract features for clips, and these features are processed to predict the actual order. The learned representations are evaluated via nearest neighbor retrieval experiments. We also use the learned networks as the pre-trained models and finetune them on the action recognition task. Three types of 3D convolutional neural networks are tested in experiments, and we gain large improvements compared to existing self-supervised methods.

Weike Jin, Zhou Zhao, Mao Gu, Jun Xiao, Furu Wei, Yueting Zhuang. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Text to speech (TTS) and automatic speech recognition (ASR) are two dual tasks in speech processing and both achieve impressive performance thanks to the recent advance in deep learning and large amount of aligned speech and text data. However, the lack of aligned data poses a major practical problem for TTS and ASR on low-resource languages. In this paper, by leveraging the dual nature of the two tasks, we propose an almost unsupervised learning method that only leverages few hundreds of paired data and extra unpaired data for TTS and ASR. Our method consists of the following components: (1) denoising auto-encoder, which reconstructs speech and text sequences respectively to develop the capability of language modeling both in speech and text domain; (2) dual transformation, where the TTS model transforms the text $y$ into speech $\hat{x}$, and the ASR model leverages the transformed pair $(\hat{x},y)$ for training, and vice versa, to boost the accuracy of the two tasks; (3) bidirectional sequence modeling, which address the error propagation problem especially in the long speech and text sequence when training with few paired data; (4) a unified model structure, which combines all the above components for TTS and ASR based on Transformer model. Our method achieves 99.84% in terms of word level intelligible rate and 2.68 MOS for TTS, and 11.7% PER for ASR on LJSpeech dataset, by leveraging only 200 paired speech and text data (about 20 minutes audio), together with extra unpaired speech and text data.

Product Quantization (PQ) has long been a mainstream for generating an exponentially large codebook at very low memory/time cost. Despite its success, PQ is still tricky for the decomposition of high-dimensional vector space, and the retraining of model is usually unavoidable when the code length changes. In this work, we propose a deep progressive quantization (DPQ) model, as an alternative to PQ, for large scale image retrieval. DPQ learns the quantization codes sequentially and approximates the original feature space progressively. Therefore, we can train the quantization codes with different code lengths simultaneously. Specifically, we first utilize the label information for guiding the learning of visual features, and then apply several quantization blocks to progressively approach the visual features. Each quantization block is designed to be a layer of a convolutional neural network, and the whole framework can be trained in an end-to-end manner. Experimental results on the benchmark datasets show that our model significantly outperforms the state-of-the-art for image retrieval. Our model is trained once for different code lengths and therefore requires less computation time. Additional ablation study demonstrates the effect of each component of our proposed model. Our code is released at https://github.com/cfm-uestc/DPQ.

Automatic question generation according to an answer within the given passage is useful for many applications, such as question answering system, dialogue system, etc. Current neural-based methods mostly take two steps which extract several important sentences based on the candidate answer through manual rules or supervised neural networks and then use an encoder-decoder framework to generate questions about these sentences. These approaches still acquire two steps and neglect the semantic relations between the answer and the context of the whole passage which is sometimes necessary for answering the question. To address this problem, we propose the Weakly Supervision Enhanced Generative Network (WeGen) which automatically discovers relevant features of the passage given the answer span in a weakly supervised manner to improve the quality of generated questions. More specifically, we devise a discriminator, Relation Guider, to capture the relations between the passage and the associated answer and then the Multi-Interaction mechanism is deployed to transfer the knowledge dynamically for our question generation system. Experiments show the effectiveness of our method in both automatic evaluations and human evaluations.

Action localization in untrimmed videos is an important topic in the field of video understanding. However, existing action localization methods are restricted to a pre-defined set of actions and cannot localize unseen activities. Thus, we consider a new task to localize unseen activities in videos via image queries, named Image-Based Activity Localization. This task faces three inherent challenges: (1) how to eliminate the influence of semantically inessential contents in image queries; (2) how to deal with the fuzzy localization of inaccurate image queries; (3) how to determine the precise boundaries of target segments. We then propose a novel self-attention interaction localizer to retrieve unseen activities in an end-to-end fashion. Specifically, we first devise a region self-attention method with relative position encoding to learn fine-grained image region representations. Then, we employ a local transformer encoder to build multi-step fusion and reasoning of image and video contents. We next adopt an order-sensitive localizer to directly retrieve the target segment. Furthermore, we construct a new dataset ActivityIBAL by reorganizing the ActivityNet dataset. The extensive experiments show the effectiveness of our method.

Open-ended video question answering aims to automatically generate the natural-language answer from referenced video contents according to the given question. Currently, most existing approaches focus on short-form video question answering with multi-modal recurrent encoder-decoder networks. Although these works have achieved promising performance, they may still be ineffectively applied to long-form video question answering due to the lack of long-range dependency modeling and the suffering from the heavy computational cost. To tackle these problems, we propose a fast hierarchical convolutional self-attention encoder-decoder network. Concretely, we first develop a hierarchical convolutional self-attention encoder to efficiently model long-form video contents, which builds the hierarchical structure for video sequences and captures question-aware long-range dependencies from video context. We then devise a multi-scale attentive decoder to incorporate multi-layer video representations for answer generation, which avoids the information missing of the top encoder layer. The extensive experiments show the effectiveness and efficiency of our method.

Code retrieval techniques and tools have been playing a key role in facilitating software developers to retrieve existing code fragments from available open-source repositories given a user query (e.g., a short natural language text describing the functionality for retrieving a particular code snippet). Despite the existing efforts in improving the effectiveness of code retrieval, there are still two main issues hindering them from being used to accurately retrieve satisfiable code fragments from large-scale repositories when answering complicated queries. First, the existing approaches only consider shallow features of source code such as method names and code tokens, but ignoring structured features such as abstract syntax trees (ASTs) and control-flow graphs (CFGs) of source code, which contains rich and well-defined semantics of source code. Second, although the deep learning-based approach performs well on the representation of source code, it lacks the explainability, making it hard to interpret the retrieval results and almost impossible to understand which features of source code contribute more to the final results. To tackle the two aforementioned issues, this paper proposes MMAN, a novel Multi-Modal Attention Network for semantic source code retrieval. A comprehensive multi-modal representation is developed for representing unstructured and structured features of source code, with one LSTM for the sequential tokens of code, a Tree-LSTM for the AST of code and a GGNN (Gated Graph Neural Network) for the CFG of code. Furthermore, a multi-modal attention fusion layer is applied to assign weights to different parts of each modality of source code and then integrate them into a single hybrid representation. Comprehensive experiments and analysis on a large-scale real-world dataset show that our proposed model can accurately retrieve code snippets and outperforms the state-of-the-art methods.

Video question answering is an important task for testing machine's ability of video understanding. The existing methods normally focus on the combination of recurrent and convolutional neural networks to capture spatial and temporal information of the video. Recently, some work has also shown that using attention mechanism can achieve better performance. In this paper, we propose a new model called Multi-interaction network for video question answering. There are two types of interactions in our model. The first type is the multi-modal interaction between the visual and textual information. The second type is the multi-level interaction inside the multi-modal interaction. Specifically, instead of using original self-attention, we propose a new attention mechanism called multi-interaction, which can capture both element-wise and segment-wise sequence interactions, simultaneously. And in addition to the normal frame-level interaction, we also take the object relations into consideration, in order to obtain more fine-grained information, such as motions and other potential relations among these objects. We evaluate our method on TGIF-QA and other two video QA datasets. The qualitative and quantitative experimental results show the effectiveness of our model, which achieves the new state-of-the-art performance.

Personalized hashtag recommendation methods aim to suggest users hashtags to annotate, categorize, and describe their posts. The hashtags, that a user provides to a post (e.g., a micro-video), are the ones which in her mind can well describe the post content where she is interested in. It means that we should consider both users' preferences on the post contents and their personal understanding on the hashtags. Most existing methods rely on modeling either the interactions between hashtags and posts or the interactions between users and hashtags for hashtag recommendation. These methods have not well explored the complicated interactions among users, hashtags, and micro-videos. In this paper, towards the personalized micro-video hashtag recommendation, we propose a Graph Convolution Network based Personalized Hashtag Recommendation (GCN-PHR) model, which leverages recently advanced GCN techniques to model the complicate interactions among and learn their representations. In our model, the users, hashtags, and micro-videos are three types of nodes in a graph and they are linked based on their direct associations. In particular, the message-passing strategy is used to learn the representation of a node (e.g., user) by aggregating the message passed from the directly linked other types of nodes (e.g., hashtag and micro-video). Because a user is often only interested in certain parts of a micro-video and a hashtag is typically used to describe the part (of a micro-video) that the user is interested in, we leverage the attention mechanism to filter the message passed from micro-videos to users and hashtags, which can significantly improve the representation capability. Extensive experiments have been conducted on two real-world micro-video datasets and demonstrate that our model outperforms the state-of-the-art approaches by a large margin.

Neural network based end-to-end text to speech (TTS) has significantly improved the quality of synthesized speech. Prominent methods (e.g., Tacotron 2) usually first generate mel-spectrogram from text, and then synthesize speech from the mel-spectrogram using vocoder such as WaveNet. Compared with traditional concatenative and statistical parametric approaches, neural network based end-to-end models suffer from slow inference speed, and the synthesized speech is usually not robust (i.e., some words are skipped or repeated) and lack of controllability (voice speed or prosody control). In this work, we propose a novel feed-forward network based on Transformer to generate mel-spectrogram in parallel for TTS. Specifically, we extract attention alignments from an encoder-decoder based teacher model for phoneme duration prediction, which is used by a length regulator to expand the source phoneme sequence to match the length of the target mel-spectrogram sequence for parallel mel-spectrogram generation. Experiments on the LJSpeech dataset show that our parallel model matches autoregressive models in terms of speech quality, nearly eliminates the problem of word skipping and repeating in particularly hard cases, and can adjust voice speed smoothly. Most importantly, compared with autoregressive Transformer TTS, our model speeds up mel-spectrogram generation by 270x and the end-to-end speech synthesis by 38x. Therefore, we call our model FastSpeech.

Video dialog is a new and challenging task, which requires an AI agent to maintain a meaningful dialog with humans in natural language about video contents. Specifically, given a video, a dialog history and a new question about the video, the agent has to combine video information with dialog history to infer the answer. And due to the complexity of video information, the methods of image dialog might be ineffectively applied directly to video dialog. In this paper, we propose a novel approach for video dialog called multi-grained convolutional self-attention context network, which combines video information with dialog history. Instead of using RNN to encode the sequence information, we design a multi-grained convolutional self-attention mechanism to capture both element and segment level interactions which contain multi-grained sequence information. Then, we design a hierarchical dialog history encoder to learn the context-aware question representation and a two-stream video encoder to learn the context-aware video representation. We evaluate our method on two large-scale datasets. Due to the flexibility and parallelism of the new attention mechanism, our method can achieve higher time efficiency, and the extensive experiments also show the effectiveness of our method.

Query-based moment retrieval aims to localize the most relevant moment in an untrimmed video according to the given natural language query. Existing works often only focus on one aspect of this emerging task, such as the query representation learning, video context modeling or multi-modal fusion, thus fail to develop a comprehensive system for further performance improvement. In this paper, we introduce a novel Cross-Modal Interaction Network (CMIN) to consider multiple crucial factors for this challenging task, including (1) the syntactic structure of natural language queries; (2) long-range semantic dependencies in video context and (3) the sufficient cross-modal interaction. Specifically, we devise a syntactic GCN to leverage the syntactic structure of queries for fine-grained representation learning, propose a multi-head self-attention to capture long-range semantic dependencies from video context, and next employ a multi-stage cross-modal interaction to explore the potential relations of video and query contents. The extensive experiments demonstrate the effectiveness of our proposed method.

Abstractive meeting summarization is a challenging problem in natural language understanding, which automatically generates the condensed summary covering the important points in the meeting conversation. However, the existing abstractive summarization works mainly focus on the structured text documents, which may be ineffectively applied to the meeting summarization task due to the lack of modeling the unstructured long-form conversational contents. In this paper, we consider the problem of abstractive meeting summarization from the viewpoint of hierarchical adaptive segmental encoder-decoder network learning. We propose the hierarchical neural encoder based on adaptive recurrent networks to learn the semantic representation of meeting conversation with adaptive conversation segmentation. We then develop the reinforced decoder network to generate the high-quality summaries for abstractive meeting summarization. We conduct the extensive experiments on the well-known AMI meeting conversation dataset to validate the effectiveness of our proposed method.

Multi-turn video question answering is a challenging task in visual information retrieval, which generates the accurate answer from the referenced video contents according to the visual conversation context and given question. However, the existing visual question answering methods mainly tackle the problem of single-turn video question answering, which may be ineffectively applied for multi-turn video question answering directly, due to the insufficiency of modeling the sequential conversation context. In this paper, we study the problem of multi-turn video question answering from the viewpoint of multi-stream hierarchical attention context reinforced network learning. We first propose the hierarchical attention context network for context-aware question understanding by modeling the hierarchically sequential conversation context structure. We then develop the multi-stream spatio-temporal attention network for learning the joint representation of the dynamic video contents and context-aware question embedding. We next devise a multi-step reasoning process to enhance the multi-stream hierarchical attention context network learning method. We finally predict the multiple-choice answer from the candidate answer set and further develop the reinforced decoder network to generate the open-ended natural language answer for multi-turn video question answering. We construct two large-scale multi-turn video question answering datasets. The extensive experiments show the effectiveness of our method.

Open-ended long-form video question answering is a challenging task in visual information retrieval, which automatically generates a natural language answer from the referenced long-form video contents according to a given question. However, the existing works mainly focus on short-form video question answering, due to the lack of modeling semantic representations from long-form video contents. In this paper, we introduce a dynamic hierarchical reinforced network for open-ended long-form video question answering, which employs an encoder-decoder architecture with a dynamic hierarchical encoder and a reinforced decoder. Concretely, we first propose a frame-level dynamic long-short term memory (LSTM) network with binary segmentation gate to learn frame-level semantic representations according to the given question. We then develop a segment-level highway LSTM network with a question-aware highway gate for segment-level semantic modeling. Furthermore, we devise the reinforced decoder with a hierarchical attention mechanism to generate natural language answers. We construct a large-scale long-form video question answering dataset. The extensive experiments on the long-form dataset and another public short-form dataset show the effectiveness of our method.

Recent artificial intelligence research has witnessed great interest in automatically generating text descriptions of images, which are known as the image captioning task. Remarkable success has been achieved on domains where a large number of paired data in multimedia are available. Nevertheless, annotating sufficient data is labor-intensive and time-consuming, establishing significant barriers for adapting the image captioning systems to new domains. In this study, we introduc a novel Multitask Learning Algorithm for cross-Domain Image Captioning (MLADIC). MLADIC is a multitask system that simultaneously optimizes two coupled objectives via a dual learning mechanism: image captioning and text-to-image synthesis, with the hope that by leveraging the correlation of the two dual tasks, we are able to enhance the image captioning performance in the target domain. Concretely, the image captioning task is trained with an encoder-decoder model (i.e., CNN-LSTM) to generate textual descriptions of the input images. The image synthesis task employs the conditional generative adversarial network (C-GAN) to synthesize plausible images based on text descriptions. In C-GAN, a generative model $G$ synthesizes plausible images given text descriptions, and a discriminative model $D$ tries to distinguish the images in training data from the generated images by $G$. The adversarial process can eventually guide $G$ to generate plausible and high-quality images. To bridge the gap between different domains, a two-step strategy is adopted in order to transfer knowledge from the source domains to the target domains. First, we pre-train the model to learn the alignment between the neural representations of images and that of text data with the sufficient labeled source domain data. Second, we fine-tune the learned model by leveraging the limited image-text pairs and unpaired data in the target domain. We conduct extensive experiments to evaluate the performance of MLADIC by using the MSCOCO as the source domain data, and using Flickr30k and Oxford-102 as the target domain data. The results demonstrate that MLADIC achieves substantially better performance than the strong competitors for the cross-domain image captioning task.

Visual Question Answering (VQA) is a challenging task that has gained increasing attention from both the computer vision and the natural language processing communities in recent years. Given a question in natural language, a VQA system is designed to automatically generate the answer according to the referenced visual content. Though there recently has been much intereset in this topic, the existing work of visual question answering mainly focuses on a single static image, which is only a small part of the dynamic and sequential visual data in the real world. As a natural extension, video question answering (VideoQA) is less explored. Because of the inherent temporal structure in the video, the approaches of ImageQA may be ineffectively applied to video question answering. In this article, we not only take the spatial and temporal dimension of video content into account but also employ an external knowledge base to improve the answering ability of the network. More specifically, we propose a knowledge-based progressive spatial-temporal attention network to tackle this problem. We obtain both objects and region features of the video frames from a region proposal network. The knowledge representation is generated by a word-level attention mechanism using the comment information of each object that is extracted from DBpedia. Then, we develop a question-knowledge-guided progressive spatial-temporal attention network to learn the joint video representation for video question answering task. We construct a large-scale video question answering dataset. The extensive experiments based on two different datasets validate the effectiveness of our method.

Multi-label community-based question classification is a challenging problem in Community-based Question Answering (CQA) services, arising in many real applications such as question navigation and expert finding. Most of the existing approaches consider the problem as content-based tag suggestion task, which suffers from the textual sparsity issue. Unlike the previous studies, we consider the problem of multi-label community-based question classification from the viewpoint of personalized sequence learning. We introduce the personalized sequence memory network that leverages not only the semantics of questions but also the personalized information of askers to provide the sequence tag learning function to capture the high-order tag dependency. The experiment on real-world dataset shows the effectiveness of our method.

Proximity search on heterogeneous graphs aims to measure the proximity between two nodes on a graph w.r.t. some semantic relation for ranking. Pioneer work often tries to measure such proximity by paths connecting the two nodes. However, paths as linear sequences have limited expressiveness for the complex network connections. In this paper, we explore a more expressive DAG (directed acyclic graph) data structure for modeling the connections between two nodes. Particularly, we are interested in learning a representation for the DAGs to encode the proximity between two nodes. We face two challenges to use DAGs, including how to efficiently generate DAGs and how to effectively learn DAG embedding for proximity search. We find distance-awareness as important for proximity search and the key to solve the above challenges. Thus we develop a novel Distance-aware DAG Embedding (D2AGE) model. We evaluate D2AGE on three benchmark data sets with six semantic relations, and we show that D2AGE outperforms the state-of-the-art baselines. We release the code on https://github.com/shuaiOKshuai.

Natural Language Inference (NLI), also known as Recognizing Textual Entailment (RTE), is one of the most important problems in natural language processing.It requires to infer the logical relationship between two given sentences.While current approaches mostly focus on the interaction architectures of the sentences, in this paper, we propose to transfer knowledge from some important discourse markers to augment the quality of the NLI model.We observe that people usually use some discourse markers such as "so" or "but" to represent the logical relationship between two sentences.These words potentially have deep connections with the meanings of the sentences, thus can be utilized to help improve the representations of them.Moreover, we use reinforcement learning to optimize a new objective function with a reward defined by the property of the NLI datasets to make full use of the labels information.Experiments show that our method achieves the state-of-the-art performance on several large-scale datasets.

Dialogue Act (DA) classification is a challenging problem in dialogue interpretation, which aims to attach semantic labels to utterances and characterize the speaker's intention. Currently, many existing approaches formulate the DA classification problem ranging from multi-classification to structured prediction, which suffer from two limitations: a) these methods are either handcrafted feature-based or have limited memories. b) adversarial examples can't be correctly classified by traditional training methods. To address these issues, in this paper we first cast the problem into a question and answering problem and proposed an improved dynamic memory networks with hierarchical pyramidal utterance encoder. Moreover, we apply adversarial training to train our proposed model. We evaluate our model on two public datasets, i.e., Switchboard dialogue act corpus and the MapTask corpus. Extensive experiments show that our proposed model is not only robust, but also achieves better performance when compared with some state-of-the-art baselines.

This study takes the lead to study the aspect/sentiment-aware abstractive review summarization in domain adaptation scenario. The proposed model CASAS (neural attentive model for Cross-domain Aspect/Sentiment-aware Abstractive review Summarization) leverages domain classification task, working on datasets of both source and target domains, to recognize the domain information of texts and transfer knowledge from source domains to target domains. The extensive experiments on Amazon reviews demonstrate that CASAS outperforms the compared methods in both out-of-domain and in-domain setups.

In this study, we explore capsule networks with dynamic routing for text classification.We propose three strategies to stabilize the dynamic routing process to alleviate the disturbance of some noise capsules which may contain "background" information or have not been successfully trained.A series of experiments are conducted with capsule networks on six text classification benchmarks.Capsule networks achieve competitive results over the compared baseline methods on 4 out of 6 datasets, which shows the effectiveness of capsule networks for text classification.We additionally show that capsule networks exhibit significant improvement when transfer single-label to multi-label text classification over the competitors.To the best of our knowledge, this is the first work that capsule networks have been empirically investigated for text modeling 1 .

Visual grounding aims to localize an object in an image referred to by a textual query phrase. Various visual grounding approaches have been proposed, and the problem can be modularized into a general framework: proposal generation, multi-modal feature representation, and proposal ranking. Of these three modules, most existing approaches focus on the latter two, with the importance of proposal generation generally neglected. In this paper, we rethink the problem of what properties make a good proposal generator. We introduce the diversity and discrimination simultaneously when generating proposals, and in doing so propose Diversified and Discriminative Proposal Networks model (DDPN). Based on the proposals generated by DDPN, we propose a high performance baseline model for visual grounding and evaluate it on four benchmark datasets. Experimental results demonstrate that our model delivers significant improvements on all the tested data-sets (e.g., 18.8% improvement on ReferItGame and 8.2% improvement on Flickr30k Entities over the existing state-of-the-arts respectively).

Conversational video question answering is a challenging task in visual information retrieval, which generates the accurate answer from the referenced video contents according to the visual conversation context and given question. However, the existing visual question answering methods mainly tackle the problem of single-turn video question answering, which may be ineffectively applied for multi-turn video question answering directly, due to the insufficiency of modeling the sequential conversation context. In this paper, we study the problem of multi-turn video question answering from the viewpoint of multi-step hierarchical attention context network learning. We first propose the hierarchical attention context network for context-aware question understanding by modeling the hierarchically sequential conversation context structure. We then develop the multi-stream spatio-temporal attention network for learning the joint representation of the dynamic video contents and context-aware question embedding. We next devise the hierarchical attention context network learning method with multi-step reasoning process for multi-turn video question answering. We construct two large-scale multi-turn video question answering datasets. The extensive experiments show the effectiveness of our method.

Retweet prediction is a challenging problem in social media sites (SMS). In this paper, we study the problem of image retweet prediction in social media, which predicts the image sharing behavior that the user reposts the image tweets from their followees. Unlike previous studies, we learn user preference ranking model from their past retweeted image tweets in SMS. We first propose heterogeneous image retweet modeling network (IRM) that exploits users' past retweeted image tweets with associated contexts, their following relations in SMS and preference of their followees. We then develop a novel attentional multi-faceted ranking network learning framework with multi-modal neural networks for the proposed heterogenous IRM network to learn the joint image tweet representations and user preference representations for prediction task. The extensive experiments on a large-scale dataset from Twitter site shows that our method achieves better performance than other state-of-the-art solutions to the problem.

In this paper, we propose a Multi-task Learning Approach for Image Captioning (MLAIC ), motivated by the fact that humans have no difficulty performing such task because they possess capabilities of multiple domains. Specifically, MLAIC consists of three key components: (i) A multi-object classification model that learns rich category-aware image representations using a CNN image encoder; (ii) A syntax generation model that learns better syntax-aware LSTM based decoder; (iii) An image captioning model that generates image descriptions in text, sharing its CNN encoder and LSTM decoder with the object classification task and the syntax generation task, respectively. In particular, the image captioning model can benefit from the additional object categorization and syntax knowledge. To verify the effectiveness of our approach, we conduct extensive experiments on MS-COCO dataset. The experimental results demonstrate that our model achieves impressive results compared to other strong competitors.

Open-ended long-form video question answering is challenging problem in visual information retrieval, which automatically generates the natural language answer from the referenced long-form video content according to the question. However, the existing video question answering works mainly focus on the short-form video question answering, due to the lack of modeling the semantic representation of long-form video contents. In this paper, we consider the problem of long-form video question answering from the viewpoint of adaptive hierarchical reinforced encoder-decoder network learning. We propose the adaptive hierarchical encoder network to learn the joint representation of the long-form video contents according to the question with adaptive video segmentation. we then develop the reinforced decoder network to generate the natural language answer for open-ended video question answering. We construct a large-scale long-form video question answering dataset. The extensive experiments show the effectiveness of our method.

Code summarization provides a high level natural language description of the function performed by code, as it can benefit the software maintenance, code categorization and retrieval. To the best of our knowledge, most state-of-the-art approaches follow an encoder-decoder framework which encodes the code into a hidden space and then decode it into natural language space, suffering from two major drawbacks: a) Their encoders only consider the sequential content of code, ignoring the tree structure which is also critical for the task of code summarization; b) Their decoders are typically trained to predict the next word by maximizing the likelihood of next ground-truth word with previous ground-truth word given. However, it is expected to generate the entire sequence from scratch at test time. This discrepancy can cause an exposure bias issue, making the learnt decoder suboptimal. In this paper, we incorporate an abstract syntax tree structure as well as sequential content of code snippets into a deep reinforcement learning framework (i.e., actor-critic network). The actor network provides the confidence of predicting the next word according to current state. On the other hand, the critic network evaluates the reward value of all possible extensions of the current state and can provide global guidance for explorations. We employ an advantage reward composed of BLEU metric to train both networks. Comprehensive experiments on a real-world dataset show the effectiveness of our proposed model when compared with some state-of-the-art methods.

Semantic proximity search on heterogeneous graph is an important task, and is useful for many applications. It aims to measure the proximity between two nodes on a heterogeneous graph w.r.t. some given semantic relation. Prior work often tries to measure the semantic proximity by paths connecting a query object and a target object. Despite the success of such path-based approaches, they often modeled the paths in a weakly coupled manner, which overlooked the rich interactions among paths. In this paper, we introduce a novel concept of interactive paths to model the inter-dependency among multiple paths between a query object and a target object. We then propose an Interactive Paths Embedding (IPE) model, which learns low-dimensional representations for the resulting interactive-paths structures for proximity estimation. We conduct experiments on seven relations with four different types of heterogeneous graphs, and show that our model outperforms the state-of-the-art baselines.

Game bots are automated programs that assist cheating users and enable them to obtain huge superiority, leading to an imbalance in the game ecosystem and the collapse of user interest. Therefore, game bot detection becomes particularly important and urgent. Among many kinds of online games, massively multiplayer online role playing games (MMORPGs), such as World of Warcraft and AION, provide immersive gaming experience and attract many loyal fans. At the same time, however, game bots in MMORPGs have proliferated in volume and method, evolving with the real-world detection methods and showing strong diversity, leaving MMORPG bot detection efforts extremely difficult. To deal with the fast-changing nature of game bots, we here proposed a generalized game bot detection framework for MMORPGs termed NGUARD, denoting NetEase Games' Guard. NGUARD takes charge of automatically differentiating game bots from humans for MMORPGs. In detail, NGUARD exploits a combination of supervised and unsupervised methods. Supervised models are utilized to detect game bots in observed patterns according to the training data. Meanwhile, unsupervised solutions are employed to detect clustered game bots and help discovering new bots. The game bot detection framework NGUARD has been implemented and deployed in multiple MMORPG productions in the NetEase Game portfolio, achieving remarkable performance improvement and acceleration compared to traditional methods. Moreover, the framework reveals outstanding robustness for game bots in mutated patterns and even in completely new patterns on account of the design of the auto-iteration mechanism.

Machine Comprehension (MC) is one of the core problems in natural language processing, requiring both understanding of the natural language and knowledge about the world. Rapid progress has been made since the release of several benchmark datasets, and recently the state-of-the-art models even surpass human performance on the well-known SQuAD evaluation. In this paper, we transfer knowledge learned from machine comprehension to the sequence-to-sequence tasks to deepen the understanding of the text. We propose MacNet: a novel encoder-decoder supplementary architecture to the widely used attention-based sequence-to-sequence models. Experiments on neural machine translation (NMT) and abstractive text summarization show that our proposed framework can significantly improve the performance of the baseline models, and our method for the abstractive text summarization achieves the state-of-the-art results on the Gigaword dataset.

In this paper, we propose a personalized dialogue generation system, which combines reinforcement learning techniques with an attention-based hierarchical recurrent encoderdecoder model. Firstly, we incorporate user-specific information into the decoder to capture user's background information and speaking style. Secondly, we employ reinforcement learning techniques to maximize future reward in dialogue, which enables our system to generate topic-coherent, informative and grammatical responses. Moreover, we propose three types of rewards to characterize good conversations. Finally, we compare the performance of the following reinforcement learning methods in dialogue generation: policy gradient, Q-learning, and actor-critic algorithms. We conduct experiments to verify the effectiveness of the proposed model on two dialogue datasets. Experimental results demonstrate that our model can generate better personalized dialogues for different users. Quantitatively, our method achieves better performance than the state-of-the-art dialogue systems in terms of BLEU score, perplexity, and human evaluation.

Dialogue Act Recognition (DAR) is a challenging problem in dialogue interpretation, which aims to associate semantic labels to utterances and characterize the speaker's intention. Currently, many existing approaches formulate the DAR problem ranging from multi-classification to structured prediction, which suffer from handcrafted feature extensions and attentive contextual dependencies. In this paper, we tackle the problem of DAR from the viewpoint of extending richer Conditional Random Field (CRF) structured dependencies without abandoning end-to-end training. We incorporate hierarchical semantic inference with memory mechanism on the utterance modeling at multiple levels. We then utilize the structured attention network on the linear-chain CRF to dynamically separate the utterances into cliques. The extensive experiments on two primary benchmark datasets Switchboard Dialogue Act (SWDA) and Meeting Recorder Dialogue Act (MRDA) datasets show that our method achieves better performance than other state-of-the-art solutions to the problem.

Semantic user search is an important task on heterogeneous social networks. Its core problem is to measure the proximity between two user objects in the network w.r.t. certain semantic user relation. State-of-the-art solutions often take a path-based approach, which uses the sequences of objects connecting a query user and a target user to measure their proximity. Despite their success, we assert that path as a low-order structure is insufficient to capture the rich semantics between two users. Therefore, in this paper we introduce a new concept of subgraph-augmented path for semantic user search. Specifically, we consider sampling a set of object paths from a query user to a target user; then in each object path, we replace the linear object sequence between its every two neighboring users with their shared subgraph instances. Such subgraph-augmented paths are expected to leverage both path»s distance awareness and subgraph»s high-order structure. As it is non-trivial to model such subgraph-augmented paths, we develop a Subgraph-augmented Path Embedding (SPE) framework to accomplish the task. We evaluate our solution on six semantic user relations in three real-world public data sets, and show that it outperforms the baselines.

Question answering is an important problem that aims to deliver specific answers to questions posed by humans in natural language. How to efficiently identify the exact answer with respect to a given question has become an active line of research. Previous approaches in factoid question answering tasks typically focus on modeling the semantic relevance or syntactic relationship between a given question and its corresponding answer. Most of these models suffer when a question contains very little content that is indicative of the answer. In this paper, we devise an architecture named the temporality-enhanced knowledge memory network (TE-KMN) and apply the model to a factoid question answering dataset from a trivia competition called quiz bowl. Unlike most of the existing approaches, our model encodes not only the content of questions and answers, but also the temporal cues in a sequence of ordered sentences which gradually remark the answer. Moreover, our model collaboratively uses external knowledge for a better understanding of a given question. The experimental results demonstrate that our method achieves better performance than several state-of-the-art methods.

We propose a new attention model for video question answering. The main idea of the attention models is to locate on the most informative parts of the visual data. The attention mechanisms are quite popular these days. However, most existing visual attention mechanisms regard the question as a whole. They ignore the word-level semantics where each word can have different attentions and some words need no attention. Neither do they consider the semantic structure of the sentences. Although the Extended Soft Attention (E-SA) model for video question answering leverages the word-level attention, it performs poorly on long question sentences. In this paper, we propose the heterogeneous tree-structured memory network (HTreeMN) for video question answering. Our proposed approach is based upon the syntax parse trees of the question sentences. The HTreeMN treats the words differently where the visual words are processed with an attention module and the verbal ones not. It also utilizes the semantic structure of the sentences by combining the neighbors based on the recursive structure of the parse trees. The understandings of the words and the videos are propagated and merged from leaves to the root. Furthermore, we build a hierarchical attention mechanism to distill the attended features. We evaluate our approach on two datasets. The experimental results show the superiority of our HTreeMN model over the other attention models especially on complex questions.

Clustering of customer transaction data is an important procedure to analyze customer behaviors in retail and e-commerce companies. Note that products from companies are often organized as a product tree, in which the leaf nodes are goods to sell, and the internal nodes (except root node) could be multiple product categories. Based on this tree, we propose the “personalized product tree”, named purchase tree, to represent a customer's transaction records. So the customers' transaction data set can be compressed into a set of purchase trees. We propose a partitional clustering algorithm, named PurTreeClust, for fast clustering of purchase trees. A new distance metric is proposed to effectively compute the distance between two purchase trees. To cluster the purchase tree data, we first rank the purchase trees as candidate representative trees with a novel separate density, and then select the top k customers as the representatives of k customer groups. Finally, the clustering results are obtained by assigning each customer to the nearest representative. We also propose a gap statistic based method to evaluate the number of clusters. A series of experiments were conducted on ten real-life transaction data sets, and experimental results show the superior performance of the proposed method.

With the rapid development of Internet movie industry social-aware movie recommendation systems (SMRs) have become a popular online web service that provide relevant movie recommendations to users. In this effort many existing movie recommendation approaches learn a user ranking model from user feedback with respect to the movie's content. Unfortunately this approach suffers from the sparsity problem inherent in SMR data. In the present work we address the sparsity problem by learning a multimodal network representation for ranking movie recommendations. We develop a heterogeneous SMR network for movie recommendation that exploits the textual description and movie-poster image of each movie as well as user ratings and social relationships. With this multimodal data we then present a heterogeneous information network learning framework called SMR-multimodal network representation learning (MNRL) for movie recommendation. To learn a ranking metric from the heterogeneous information network we also developed a multimodal neural network model. We evaluated this model on a large-scale dataset from a real world SMR Web site and we find that SMR-MNRL achieves better performance than other state-of-the-art solutions to the problem.

Many real-world networks have a rich collection of objects. The semantics of these objects allows us to capture different classes of proximities, thus enabling an important task of semantic proximity search. As the core of semantic proximity search, we have to measure the proximity on a heterogeneous graph, whose nodes are various types of objects. Most of the existing methods rely on engineering features about the graph structure between two nodes to measure their proximity. With recent development on graph embedding, we see a good chance to avoid feature engineering for semantic proximity search. There is very little work on using graph embedding for semantic proximity search. We also observe that graph embedding methods typically focus on embedding nodes, which is an "indirect'' approach to learn the proximity. Thus, we introduce a new concept of proximity embedding, which directly embeds the network structure between two possibly distant nodes. We also design our proximity embedding, so as to flexibly support both symmetric and asymmetric proximities. Based on the proximity embedding, we can easily estimate the proximity score between two nodes and enable search on the graph. We evaluate our proximity embedding method on three real-world public data sets, and show it outperforms the state-of-the-art baselines.

Nowadays the community-based question answering (CQA) sites become the popular Internet-based web service, which have accumulated millions of questions and their posted answers over time. Thus, question answering becomes an essential problem in CQA sites, which ranks the high-quality answers to the given question. Currently, most of the existing works study the problem of question answering based on the deep semantic matching model to rank the answers based on their semantic relevance, while ignoring the authority of answerers to the given question. In this paper, we consider the problem of community-based question answering from the viewpoint of asymmetric multi-faceted ranking network embedding. We propose a novel asymmetric multi-faceted ranking network learning framework for community-based question answering by jointly exploiting the deep semantic relevance between question-answer pairs and the answerers' authority to the given question. We then develop an asymmetric ranking network learning method with deep recurrent neural networks by integrating both answers' relative quality rank to the given question and the answerers' following relations in CQA sites. The extensive experiments on a large-scale dataset from a real world CQA site show that our method achieves better performance than other state-of-the-art solutions to the problem.

Machine Reading and Comprehension recently has drawn a fair amount of attention in the field of natural language processing. In this paper, we consider integrating side information to improve machine comprehension on answering cloze-style questions more precisely. To leverage the external information, we present a novel attention-based architecture which could feed the side information representations into word level embeddings to explore the comprehension performance. Our experiments show consistent improvements of our model over various baselines.

Recent AI research has witnessed increasing interests in automatically generating image descriptions in text, which is coined as theimage captioning problem. Significant progresses have been made in domains where plenty of labeled training data (i.e. image-text pairs) are readily available or collected. However, obtaining rich annotated data is a time-consuming and expensive process, creating a substantial barrier for applying image captioning methods to a new domain. In this paper, we propose a cross-domain image captioning approach that uses a novel dual learning mechanism to overcome this barrier. First, we model the alignment between the neural representations of images and that of natural languages in the source domain where one can access sufficient labeled data. Second, we adjust the pre-trained model based on examining limited data (or unpaired data) in the target domain. In particular, we introduce a dual learning mechanism with a policy gradient method that generates highly rewarded captions. The mechanism simultaneously optimizes two coupled objectives: generating image descriptions in text and generating plausible images from text descriptions, with the hope that by explicitly exploiting their coupled relation, one can safeguard the performance of image captioning in the target domain. To verify the effectiveness of our model, we use MSCOCO dataset as the source domain and two other datasets (Oxford-102 and Flickr30k) as the target domains. The experimental results show that our model consistently outperforms previous methods for cross-domain image captioning.

We study the problem of identifying the topics and sentiments and tracking their shifts from social media texts in different geographical regions during emergencies and disasters.We propose a location-based dynamic sentiment-topic model (LDST) which can jointly model topic, sentiment, time and Geolocation information.The experimental results demonstrate that LDST performs very well at discovering topics and sentiments from social media and tracking their shifts in different geographical regions during emergencies and disasters 1 .

Link prediction is a challenging problem for complex network analysis, arising in many disciplines such as social networks and telecommunication networks. Currently, many existing approaches estimate the proximity of the link endpoints for link prediction from their feature or the local neighborhood around them, which suffer from the localized view of network connections and insufficiency of discriminative feature representation. In this paper, we consider the problem of link prediction from the viewpoint of learning discriminative path-based proximity ranking metric embedding. We propose a novel ranking metric network learning framework by jointly exploiting both node-level and path-level attentional proximity of the endpoints for link prediction. We then develop the path-based dual-level reasoning attentional learning method with recurrent neural network for proximity ranking metric embedding. The extensive experiments on two large-scale datasets show that our method achieves better performance than other state-of-the-art solutions to the problem.

Microblog Sentiment Classification (MSC) is a challenging task in microblog mining, arising in many applications such as stock price prediction and crisis management. Currently, most of the existing approaches learn the user sentiment model from their posted tweets in microblogs, which suffer from the insufficiency of discriminative tweet representation. In this paper, we consider the problem of microblog sentiment classification from the viewpoint of heterogeneous MSC network embedding. We propose a novel recurrent random walk network learning framework for the problem by exploiting both users’ posted tweets and their social relations in microblogs. We then introduce the deep recurrent neural networks with random-walk layer for heterogeneous MSC network embedding, which can be trained end-to-end from the scratch. Weemploytheback-propagationmethodfortraining the proposed recurrent random walk network model. The extensive experiments on the large-scale public datasets from Twitter show that our method achieves better performance than other state-of-the-art solutions to the problem.

Open-ended video question answering is a challenging problem in visual information retrieval, which automatically generates the natural language answer from the referenced video content according to the question. However, the existing visual question answering works only focus on the static image, which may be ineffectively applied to video question answering due to the temporal dynamics of video contents. In this paper, we consider the problem of open-ended video question answering from the viewpoint of spatio-temporal attentional encoder-decoder learning framework. We propose the hierarchical spatio-temporal attention network for learning the joint representation of the dynamic video contents according to the given question. We then develop the encoder-decoder learning method with reasoning recurrent neural networks for open-ended video question answering. We construct a large-scale video question answering dataset. The extensive experiments show the effectiveness of our method.

Understanding the semantics in videos is a complex but crucial task in video analysis. This paper focuses on localizing category-independent events, actions or other semantics in an untrimmed video, referred as salient temporal proposal localization. Traditional methods like sliding window have a high computational cost due to the densely sampling of different video segments. We propose a reinforcement learning based method, which trains a localizer that learns a search policy that, instead of exploring every video segment, finds an optimal search path to locate a salient proposal based on the currently observing video segment in a tree structure, therefore reduces the number of video segments fed into the proposal detector. In each search step, a localizer is trained to iteratively select the next sub-region containing salient proposals to continue the search, and a proposal detector is trained to recognize salient proposal from the sub-regions. The experiments demonstrate that our method is able to precisely detect salient proposals with a comparable recall and with much fewer candidate windows.

Recently image question answering (ImageQA) has gained lots of attention in the research community. However, as its natural extension, video question answering (VideoQA) is less explored. Although both tasks look similar, VideoQA is more challenging mainly because of the complexity and diversity of videos. As such, simply extending the ImageQA methods to videos is insufficient and suboptimal. Particularly, working with the video needs to model its inherent temporal structure and analyze the diverse information it contains. In this paper, we consider exploiting the appearance and motion information resided in the video with a novel attention mechanism. More specifically, we propose an end-to-end model which gradually refines its attention over the appearance and motion features of the video using the question as guidance. The question is processed word by word until the model generates the final optimized attention. The weighted representation of the video, as well as other contextual information, are used to generate the answer. Extensive experiments show the advantages of our model compared to other baseline models. We also demonstrate the effectiveness of our model by analyzing the refined attention weights during the question answering procedure.

Video question answering is a challenging task in visual information retrieval, which provides the accurate answer from the referenced video contents according to the given question. However, the existing visual question answering approaches mainly tackle the problem of static image question answering, which may be ineffectively applied for video question answering directly, due to the insufficiency of modeling the video temporal dynamics. In this paper, we study the problem of video question answering from the viewpoint of hierarchical dual-level attention network learning. We obtain the object appearance and movement information in the video based on both frame-level and segment-level feature representation methods. We then develop the hierarchical duallevel attention networks to learn the question-aware video representations with word-level and question-level attention mechanisms. We next devise the question-level fusion attention mechanism for our proposed networks to learn the questionaware joint video representation for video question answering. We construct two large-scale video question answering datasets. The extensive experiments validate the effectiveness of our method.

In this paper, we propose a novel personalized response generation model via domain adaptation (PRG-DM). First, we learn the human responding style from large general data (without user-specific information). Second, we fine tune the model on a small size of personalized data to generate personalized responses with a dual learning mechanism. Moreover, we propose three new rewards to characterize good conversations that are personalized, informative and grammatical. We employ the policy gradient method to generate highly rewarded responses. Experimental results show that our model can generate better personalized responses for different users.

Video Question Answering is a challenging problem in visual information retrieval, which provides the answer to the referenced video content according to the question. However, the existing visual question answering approaches mainly tackle the problem of static image question, which may be ineffectively for video question answering due to the insufficiency of modeling the temporal dynamics of video contents. In this paper, we study the problem of video question answering by modeling its temporal dynamics with frame-level attention mechanism. We propose the attribute-augmented attention network learning framework that enables the joint frame-level attribute detection and unified video representation learning for video question answering. We then incorporate the multi-step reasoning process for our proposed attention network to further improve the performance. We construct a large-scale video question answering dataset. We conduct the experiments on both multiple-choice and open-ended video question answering tasks to show the effectiveness of the proposed method.

With the rapid development of mobile devices, point-of-interest (POI) suggestion has become a popular online web service, which provides attractive and interesting locations to users. In order to provide interesting POIs, many existing POI recommendation works learn the latent representations of users and POIs from users' past visiting POIs, which suffers from the sparsity problem of POI data. In this paper, we consider the problem of POI suggestion from the viewpoint of learning geosocial multimedia network representations. We propose a novel max-margin metric geosocial multimedia network representation learning framework by exploiting users' check-in behavior and their social relations. We then develop a random-walk based learning method with max-margin metric network embedding. We evaluate the performance of our method on a large-scale geosocial multimedia network dataset and show that our method achieves the best performance than other state-of-the-art solutions.

Community question answering(CQA) services have arisen as a popular knowledge sharing pattern for netizens. With abundant interactions among users, individuals are capable of obtaining satisfactory information. However, it is not effective for users to attain satisfying answers within minutes. Users have to check the progress over time until the appropriate answers submitted. We address this problem as a user personalized satisfaction prediction task. Existing methods usually exploit manual feature selection. It is not desirable as it requires careful design and is labor intensive. In this paper, we settle this issue by developing a new multiple instance deep learning framework. Specifically, in our settings, each question follows a multiple instance learning assumption, where its obtained answers can be regarded as instance sets in a bag and we define the question resolved with at least one satisfactory answer. We design an efficient framework exploiting multiple instance learning property with deep learning tactic to model the question-answer pairs relevance and rank the asker's satisfaction possibility. Extensive experiments on large-scale datasets from different forums of Stack Exchange demonstrate the feasibility of our proposed framework in predicting asker personalized satisfaction.

Video question answering is an important task toward scene understanding and visual data retrieval. However, current visual question answering works mainly focus on a single static image, which is distinct from the dynamic and sequential visual data in the real world. Their approaches cannot utilize the temporal information in videos. In this paper, we introduce the task of free-form open-ended video question answering. The open-ended answers enable wider applications compared with the common multiple-choice tasks in Visual-QA. We first propose a data set for open-ended Video-QA with the automatic question generation approaches. Then, we propose our sequential video attention and temporal question attention models. These two models apply the attention mechanism on videos and questions, while preserving the sequential and temporal structures of the guides. The two models are integrated into the model of unified attention. After the video and the question are encoded, the answers are generated wordwisely from our models by a decoder. In the end, we evaluate our models on the proposed data set. The experimental results demonstrate the effectiveness of our proposed model.

During the last decade, community-based question answering (CQA) sites have accumulated a vast amount of questions and their crowdsourced answers over time. How to efficiently identify the quality of answers that are relevant to a given question has become an active line of research in CQA. The major challenge of CQA is the accurate selection of high-quality answers w.r.t given questions. Previous approaches tend to model the semantic matching between individual pair of one question and its corresponding answer (how fitting an answer is to a posted question). However, these works ignore the temporal interactions between answers (how previous answers influence the late posted answers). For example, a rational user likely adapts others' opinions, revises his inclinations, and posts a more appropriate answer after understanding the given question and previously posted answers. As a result, this paper devises an architecture named Temporal Interaction and Causal Influence LSTM (TC-LSTM) to effectively leverage not only the causal influence between question-answer (how appropriate an answer is for a given question) but also the temporal interactions between answers-answer (how a high-quality answer gradually forms). In particular, long short-term memory (LSTM) is used to capture the explicit question-answer influence and the implicit answers-answer interactions. Experiments are conducted on SemEval 2015 CQA dataset for answer classification task and Baidu Zhidao Dataset for answer ranking task. The experimental results show the advantage of our model comparing with other state-of-the-art methods.

Community-based question answering (cQA) sites have accumulated vast amount of questions and corresponding crowdsourced answers over time. How to efficiently share the underlying information and knowledge from reliable (usually highly-reputable) answerers has become an increasingly popular research topic. A major challenge in cQA tasks is the accurate matching of high-quality answers w.r.t given questions. Many of traditional approaches likely recommend corresponding answers merely depending on the content similarity between questions and answers, therefore suffer from the sparsity bottleneck of cQA data. In this paper, we propose a novel framework which encodes not only the contents of question-answer(Q-A) but also the social interaction cues in the community to boost the cQA tasks. More specifically, our framework collaboratively utilizes the rich interaction among questions, answers and answerers to learn the relative quality rank of different answers w.r.t a same question. Moreover, the information in heterogeneous social networks is comprehensively employed to enhance the quality of question-answering (QA) matching by our deep random walk learning framework. Extensive experiments on a large-scale dataset from a real world cQA site show that leveraging the heterogeneous social information indeed achieves better performance than other state-of-the-art cQA methods.

Multi-modal sparse coding has played an important role in many multimedia applications, where data are usually with multiple modalities. Recently, various multi-modal sparse coding approaches have been proposed to learn sparse codes of multi-modal data, which assume that data appear in all modalities, or at least there is one modality containing all data. However, in real applications, it is often the case that some modalities of the data may suffer from missing information and thus result in partial multi-modality data. In this paper, we propose to solve the partial multi-modal sparse coding problem via multi-modal similarity structure regularization. Specifically, we propose a partial multi-modal sparse coding framework termed Adaptive Partial Multi-Modal Similarity Structure Regularization for Sparse Coding (AdaPM2SC), which preserves the similarity structure within the same modality and between different modalities. Experimental results conducted on two real-world datasets demonstrate that AdaPM2SC significantly outperforms the state-of-the-art methods under partial multi-modality scenario.

Given a complex task requiring a specific set of skills, it is useful to form a team of experts who work in a collaborative manner against time and many different costs. Team formation in the setting of social networks is a fundamental problem in many database or web applications. For example, we need to find a suitable team to answer community-based questions and collaborative software development. It is also well-recognized that forming a suitable team in social networks is non-trivial, since the problem involves many cost factors such as communication overhead and load balancing. Although many algorithms have been yet proposed for resolving this problem, most of them are based on very different criteria, and performance metrics, and their performance has not been empirically compared. In this paper, we first compare and contrast all the state-of-the-art team formation algorithms. Next, we propose a benchmark that enables fair comparison amongst these algorithms. We then implement these algorithms using a common platform called the Unified System for Team Formation (USTF) and evaluate their performance using several real datasets. We also present a case study that shows the performance of different algorithms in a range of real world cases. All our experiments are repeatable and the code and the datasets are publicly accessible for further research[40].

Feature selection is a challenging problem for high dimensional data processing, which arises in many real applications such as data mining, information retrieval, and pattern recognition. In this paper, we study the problem of unsupervised feature selection. The problem is challenging due to the lack of label information to guide feature selection. We formulate the problem of unsupervised feature selection from the viewpoint of graph regularized data reconstruction. The underlying idea is that the selected features not only preserve the local structure of the original data space via graph regularization, but also approximately reconstruct each data point via linear combination. Therefore, the graph regularized data reconstruction error becomes a natural criterion for measuring the quality of the selected features. By minimizing the reconstruction error, we are able to select the features that best preserve both the similarity and discriminant information in the original data. We then develop an efficient gradient algorithm to solve the corresponding optimization problem. We evaluate the performance of our proposed algorithm on text clustering. The extensive experiments demonstrate the effectiveness of our proposed approach.

A social recommendation system has attracted a lot of attention recently in the research communities of information retrieval, machine learning, and data mining. Traditional social recommendation algorithms are often based on batch machine learning methods which suffer from several critical limitations, e.g., extremely expensive model retraining cost whenever new user ratings arrive, unable to capture the change of user preferences over time. Therefore, it is important to make social recommendation system suitable for real-world online applications where data often arrives sequentially and user preferences may change dynamically and rapidly. In this paper, we present a new framework of online social recommendation from the viewpoint of online graph regularized user preference learning (OGRPL), which incorporates both collaborative user-item relationship as well as item content features into an unified preference learning process. We further develop an efficient iterative procedure, OGRPL-FW which utilizes the Frank-Wolfe algorithm, to solve the proposed online optimization problem. We conduct extensive experiments on several large-scale datasets, in which the encouraging results demonstrate that the proposed algorithms obtain significantly lower errors (in terms of both RMSE and MAE) than the state-of-the-art online recommendation methods when receiving the same amount of training data in the online learning process.

Crowd-selection is essential to crowdsourcing applications, since choosing the right workers with particular expertise to carry out specific crowdsourced tasks is extremely important. The central problem is simple but tricky: given a crowdsourced task, who is the right worker to ask? Currently, most existing work has mainly studied the problem of crowd-selection for simple crowdsourced tasks such as decision making and sentiment analysis. Their crowd-selection procedures are based on the trustworthiness of workers. However, for some complex tasks such as document review and question answering, selecting workers based on the latent category of tasks is a better solution. In this paper, we formulate a new problem of task-driven crowd-selection for complex tasks. We first develop a Bayesian generative model to exploit "who knows what" for the workers in the crowdsourcing environment. The model provides a principle and natural framework for capturing the latent skills of workers as well as the latent categories of crowdsourced tasks. The inference of the latent skills of workers is based on past resolved crowdsourced tasks with feedback scores. We assume that the feedback scores can illustrate the performance of the workers for the tasks. We then devise a variational algorithm that transforms the latent skill inference with the proposed model into a standard optimization problem, which can be solved efficiently. We verify the performance of our method through extensive experiments on the data collected from three well-known crowdsourcing platforms for question answering tasks such as Quora, Yahoo! Answer and Stack Overflow.

Cross-modal retrieval is a very hot research topic that is imperative to many applications involving multi-modal data. Discovering an appropriate representation for multi-modal data and learning a ranking function are essential to boost the cross-media retrieval. Motivated by the assumption that a compositional cross-modal semantic representation (pairs of images and text) is more attractive for cross-modal ranking, this paper exploits the existing image-text databases to optimize a ranking function for cross-modal retrieval, called deep compositional cross-modal learning to rank (C2MLR). In this paper, C2MLR considers learning a multi-modal embedding from the perspective of optimizing a pairwise ranking problem while enhancing both local alignment and global alignment. In particular, the local alignment (i.e., the alignment of visual objects and textual words) and the global alignment (i.e., the image-level and sentence-level alignment) are collaboratively utilized to learn the multi-modal embedding common space in a max-margin learning to rank manner. The experiments demonstrate the superiority of our proposed C2MLR due to its nature of multi-modal compositional embedding.

The convex hull of a set of two-dimensional points, P, is the minimal convex polygon that contains all the points in P. Convex hull is important in many applications such as GIS, statistical analysis and data mining. Due to the ubiquity of data uncertainty such as location uncertainty in real-world applications, we study the concept of convex hull over uncertain data in 2D space. We propose the Probabilistic Convex Hull(PCH) query and demonstrate its applications, such as Flickr landscape photo extraction and activity region visualization, where location uncertainty is incurred by GPS devices or sensors. To tackle the problem of possible world explosion, we develop an O(N3) algorithm based on geometric properties, where N is the data size. We further improve this algorithm with spatial indices and effective pruning techniques, which prune the majority of data instances. To achieve better time complexity, we propose another O(N2log N) algorithm, by maintaining a probability oracle in the form of a circular array with nice properties. Finally, to support applications that require fast response, we develop a Gibbs-sampling-based approximation algorithm which efficiently finds the PCH with high accuracy. Extensive experiments are conducted to verify the efficiency of our algorithms for answering PCH queries.

Expert finding for question answering is a challenging problem in community-based question answering (CQA) systems, arising in many real applications such as question routing and identification of best answers. In order to provide high-quality experts, many existing approaches learn the user model from their past question-answering activities in CQA systems. However, the past activities of users in most CQA systems are rather few, and thus the user model may not be well inferred in practice. In this paper, we consider the problem of expert finding from the viewpoint of missing value estimation. We then employ users' social networks for inferring user model, and thus improve the performance of expert finding in CQA systems. In addition, we develop a novel graph-regularized matrix completion algorithm for inferring the user model. We further develop two efficient iterative procedures, GRMC-EGM and GRMC-AGM, to solve the optimization problem. GRMC-EGM utilizes the Extended Gradient Method (EGM), while GRMC-AGM applies the Accelerated proximal Gradient search Method (AGM), for the optimization. We evaluate our methods on the well-known question answering system Quora, and the popular social network Twitter. Our empirical study shows the effectiveness of the proposed algorithms in comparison to the state-of-the-art expert finding algorithms.

Truth discovery is a long-standing problem for assessing the validity of information from various data sources that may provide different and conflicting information. With the increasing prominence of data streams arising in a wide range of applications such as weather forecast and stock price prediction, effective techniques for truth discovery in data streams are demanded. However, existing work mainly focuses on truth discovery in the context of static databases, which is not applicable in applications involving streaming data. This motivates us to develop new techniques to tackle the problem of truth discovery in data streams.

An essential component of building a successful crowdsourcing market is effective task matching, which matches a given task to the right crowdworkers. In order to provide high- quality task matching, crowdsourcing systems rely on past task-solving activities of crowdworkers. However, the average number of past activities of crowdworkers in most crowd- sourcing systems is very small. We call the workers who have only solved a small number of tasks cold-start crowdworkers. We observe that most of the workers in crowdsourcing systems are cold-start crowdworkers, and crowdsourcing systems actually enjoy great benefits from cold-start crowd-workers. However, the problem of task matching with the presence of many cold-start crowdworkers has not been well studied. We propose a new approach to address this issue. Our main idea, motivated by the prevalence of online social networks, is to transfer the knowledge about crowdworkers in their social networks to crowdsourcing systems for task matching. We propose a SocialTransfer model for cold-start crowdsourcing, which not only infers the expertise of warm- start crowdworkers from their past activities, but also transfers the expertise knowledge to cold-start crowdworkers via social connections. We evaluate the SocialTransfer model on the well-known crowdsourcing system Quora, using knowledge from the popular social network Twitter. Experimental results show that, by transferring social knowledge, our method achieves significant improvements over the state-of-the-art methods.

Data uncertainty is inherent in many real-world applications such as environmental surveillance and mobile tracking. Mining sequential patterns from inaccurate data, such as those data arising from sensor readings and GPS trajectories, is important for discovering hidden knowledge in such applications. In this paper, we propose to measure pattern frequentness based on the possible world semantics. We establish two uncertain sequence data models abstracted from many real-life applications involving uncertain sequence data, and formulate the problem of mining probabilistically frequent sequential patterns (or p-FSPs) from data that conform to our models. However, the number of possible worlds is extremely large, which makes the mining prohibitively expensive. Inspired by the famous PrefixSpan algorithm, we develop two new algorithms, collectively called U-PrefixSpan, for p-FSP mining. U-PrefixSpan effectively avoids the problem of “possible worlds explosion”, and when combined with our four pruning and validating methods, achieves even better performance. We also propose a fast validating method to further speed up our U-PrefixSpan algorithm. The efficiency and effectiveness of U-PrefixSpan are verified through extensive experiments on both real and synthetic datasets.

Databases often offer poor answers with respect to judgemental queries such as asking the best among the movies shown in recent months. Processing such queries requires human input for providing missing information in order to clarify uncertainty or inconsistency in queries. Nowadays, it is common to see people seeking answers on micro-blogs through asking or sharing questions with their friends. This can be easily done via smart phones, which diffuse a question to a large number of users through message propagation in microblogs. This trend is important and known as CrowdSearch. Due to conflicting attitudes among crowds, the majority vote is employed as a crowd-wisdom aggregation schema. In this demo, we show the problem of minimizing the monetary cost of a crowdsourced query, given the specified expected accuracy of the aggregated answer. We present CrowdSeed, a system that automatically integrates human input for processing queries imposed on microblogs. We demonstrate the effectiveness and efficiency of our system using real world data, as well as presenting interesting results from a game called "Who is in the CrowdSeed?".

Crowd selection is essential to crowd sourcing applications, since choosing the right workers with particular expertise to carry out crowdsourced tasks is extremely important. The central problem is simple but tricky: given a crowdsourced task, who are the most knowledgable users to ask? In this demo, we show our framework that tackles the problem of crowdsourced task assignment on Twitter according to the social activities of its users. Since user profiles on Twitter do not reveal user interests and skills, we transfer the knowledge from categorized Yahoo! Answers datasets for learning user expertise. Then, we select the right crowd for certain tasks based on user expertise. We study the effectiveness of our system using extensive user evaluation. We further engage the attendees to participate a game called--Whom to Ask on Twitter?. This helps understand our ideas in an interactive manner. Our crowd selection can be accessed by the following url http://webproject2.cse.ust.hk:8034/tcrowd/.

RFID (radio frequency identification) technology has been widely used for object tracking in many real-life applications, such as inventory monitoring and product flow tracking. These applications usually rely on passive RFID technologies rather than active ones, since passive RFID tags are more attractive than active ones in many aspects, such as lower tag cost and simpler maintenance.

Finding reverse nearest neighbors (RNNs) is an important operation in spatial databases. The problem of evaluating RNN queries has already received considerable attention due to its importance in many real-world applications, such as resource allocation and disaster response. While RNN query processing has been extensively studied in Euclidean space, no work ever studies this problem on land surfaces. However, practical applications of RNN queries involve terrain surfaces that constrain object movements, which rendering the existing algorithms inapplicable.

In recent years, RFID technologies have been used in many applications, such as inventory checking and object tracking. However, raw RFID data are inherently unreliable due to physical device limitations and different kinds of environmental noise. Currently, existing work mainly focuses on RFID data cleansing in a static environment (e.g. inventory checking). It is therefore difficult to cleanse RFID data streams in a mobile environment (e.g. object tracking) using the existing solutions, which do not address the data missing issue effectively.

Data uncertainty is inherent in many real-world applications such as environmental surveillance and mobile tracking. As a result, mining sequential patterns from inaccurate data, such as sensor readings and GPS trajectories, is important for discovering hidden knowledge in such applications. Previous work uses expected support as the measurement of pattern frequentness, which has inherent weaknesses with respect to the underlying probability model, and is therefore ineffective for mining high-quality sequential patterns from uncertain sequence databases.

Uncertain data is inherently important in a lot of real-world applications, such as environmental surveillance and mobile tracking. Probabilistic convex hull is very useful for discovering the territory of imprecise data in such applications with a high confidence. In order to deal with this, we propose and study probabilistic convex hull queries based on the possible world semantics, which are able to retrieve the objects whose probability of being on the convex hull is at least α. The demonstration is based on animal tracking whose GPS coordinate is no longer considered to be precise due to device limitation or privacy issues. We demonstrate two interesting results from studying the migration habit of one specific species and the correlation between species through probabilistic convex hull queries.