Massive multi-modality datasets are fundamental to the success of large video-language models. However, existing datasets often focus on providing textual descriptions for visual content, treating audio, particularly music, as weakly related information. This overlooks the inherent semantic correlation between visual narratives and musical scores, limiting the development of models for fine-grained cross-modal understanding and generation. To address this gap, we introduce VMChill, a large-scale, fine-grained multimodal video dataset. We leverage trailers as our data source, as they are professionally edited to create a strong synergy between visual pacing, scene transitions, and background music for narrative and emotional impact. Our dataset comprises over 20 million video clips derived from more than 27.1k hours of high-resolution trailer videos. To annotate this data, we propose a systematic multimodal captioning framework. This framework first employs specialized unimodal models to extract descriptive features from multiple perspectives, including visual content, motion dynamics, and musical attributes (e.g., genre, instruments, mood). Subsequently, a large language model (LLM) is utilized to adaptively fuse these diverse descriptions into a single, coherent, and rich multimodal caption. This process yields VMChill-2M, a high-quality subset of 2 million clips with detailed multimodal annotations, and VMChill-Test, a manually refined test set for evaluation. We conduct extensive experiments on downstream tasks, including video understanding and generation, to establish benchmarks and demonstrate the dataset's quality. The results validate that VMChill effectively enhances model performance, highlighting its potential to facilitate future research in fine-grained multimodal learning. We will release the dataset, annotation codebase, and processing pipelines to support community research.

Diffusion models have demonstrated remarkable success in generative tasks, including audio super-resolution (SR). In many applications like movie post-production and album mastering, substantial computational budgets are available for achieving superior audio quality. However, while existing diffusion approaches typically increase sampling steps to improve quality, the performance remains fundamentally limited by the stochastic nature of the sampling process, leading to high-variance and quality-limited outputs. Here, rather than simply increasing the number of sampling steps, we propose a different paradigm through inference-time scaling for SR, which explores multiple solution trajectories during the sampling process. Different task-specific verifiers are developed, and two search algorithms, including the random search and zero-order search for SR, are introduced. By actively guiding the exploration of the high-dimensional solution space through verifier-algorithm combinations, we enable more robust and higher-quality outputs. Through extensive validation across diverse audio domains (speech, music, sound effects) and frequency ranges, we demonstrate consistent performance gains, achieving improvements of up to 9.70% in aesthetics, 5.88% in speaker similarity, 15.20% in word error rate, and 46.98% in spectral distance for speech SR from 4 kHz to 24 kHz, showcasing the effectiveness of our approach.

The development of speech understanding and generation has been significantly accelerated by the availability of large-scale, high-quality speech datasets. Among these, ASR and TTS are regarded as the most established and fundamental tasks. However, for Cantonese (Yue Chinese), spoken by approximately 84.9 million native speakers worldwide, limited annotated resources have hindered progress and resulted in suboptimal ASR and TTS performance. To address this challenge, we propose WenetSpeech-Pipe, an integrated pipeline for building large-scale speech corpus with multi-dimensional annotation tailored for speech understanding and generation. Based on this pipeline, we release WenetSpeech-Yue, the first large-scale Cantonese speech corpus with multi-dimensional annotation for ASR and TTS, covering 21,800 hours across 10 domains with annotations including ASR transcription, text confidence, speaker identity, age, gender, speech quality scores, among other annotations. We also release WSYue-eval, a comprehensive Cantonese benchmark with two components: WSYue-ASR-eval, a manually annotated set for evaluating ASR on short and long utterances, code-switching, and diverse acoustic conditions, and WSYue-TTS-eval, with base and coverage subsets for standard and generalization testing. Experimental results show that models trained on WenetSpeech-Yue achieve competitive results against state-of-the-art (SOTA) Cantonese ASR and TTS systems, including commercial and LLM-based models, highlighting the value of our dataset and pipeline.

Chi-Min Chan, Han Zhu, Chunyang Jiang, Jiaming Ji, Juntao Dai, Wei Xue, Sirui Han, Yike Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Chi-Min Chan, Yujin Zhou, Pengcheng Wen, Boqin Yin, Jiaming Ji, Juntao Dai, Wei Xue, Sirui Han, Yike Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Abstract The potential for higher-resolution image generation using pretrained diffusion models is immense. However, these models often struggle with object repetition and structural artifacts especially when scaling to 4K resolution and beyond. Our analysis reveals that causes the problem, a single prompt for the generation of multiple scales provides insufficient efficacy. To address this, we propose HiPrompt, a new tuning-free solution that tackles the above problems by introducing hierarchical prompts. The hierarchical prompts provide both global and local semantic guidance. Specifically, the global prompt captures overall scene semantics from user input, while local guidance comes from patch-wise descriptions generated by MLLMs to refine regional structures and textures. Furthermore, during inverse denoising, noise is decomposed into low- and high-frequency components, each conditioned on different prompt levels, facilitating prompt-guided denoising under hierarchical semantic guidance. It further allows the generation to focus more on local spatial regions and ensures the generated images maintain coherent local and global semantics, structures, and textures with high definition. Extensive experiments demonstrate that HiPrompt outperforms state-of-the-art works in higher-resolution image generation, significantly reducing object repetition and enhancing structural quality. The demo and code can be found on the project website: https://liuxinyv.github.io/HiPrompt/ .

Large language models (LLMs) have shown remarkable capability in numerous tasks and applications. However, fine-tuning LLMs using high-quality datasets under external supervision remains prohibitively expensive. In response, LLM self-improvement approaches have been vibrantly developed recently. The typical paradigm of LLM self-improvement involves training LLM on self-generated data, part of which may be detrimental and should be filtered out due to the unstable data quality. While current works primarily employs filtering strategies based on answer correctness, in this paper, we demonstrate that filtering out correct but with high distribution shift extent (DSE) samples could also benefit the results of self-improvement. Given that the actual sample distribution is usually inaccessible, we propose a new metric called DS weight to approximate DSE, inspired by the Importance Weighting methods. Consequently, we integrate DS weight with self-consistency to comprehensively filter the self-generated samples and fine-tune the language model. Experiments show that with only a tiny valid set (up to 5% size of the training set) to compute DS weight, our approach can notably promote the reasoning ability of current LLM self-improvement methods. The resulting performance is on par with methods that rely on external supervision from pre-trained reward models.

Recent advancements in audio generation have been significantly propelled by the capabilities of Large Language Models (LLMs). The existing research on audio LLM has primarily focused on enhancing the architecture and scale of audio language models, as well as leveraging larger datasets, and generally, acoustic codecs, such as EnCodec, are used for audio tokenization. However, these codecs were originally designed for audio compression, which may lead to suboptimal performance in the context of audio LLM. Our research aims to address the shortcomings of current audio LLM codecs, particularly their challenges in maintaining semantic integrity in generated audio. For instance, existing methods like VALL-E, which condition acoustic token generation on text transcriptions, often suffer from content inaccuracies and elevated word error rates (WER) due to semantic misinterpretations of acoustic tokens, resulting in word skipping and errors. To overcome these issues, we propose a straightforward yet effective approach called X-Codec. X-Codec incorporates semantic features from a pre-trained semantic encoder before the Residual Vector Quantization (RVQ) stage and introduces a semantic reconstruction loss after RVQ. By enhancing the semantic ability of the codec, X-Codec significantly reduces WER in speech synthesis tasks and extends these benefits to non-speech applications, including music and sound generation. Our experiments in text-to-speech, music continuation, and text-to-sound tasks demonstrate that integrating semantic information substantially improves the overall performance of language models in audio generation.

Boyi Kang, Xinfa Zhu, Zihan Zhang, Zhen Ye, Mingshuai Liu, Ziqian Wang, Yike Zhu, Guobin Ma, Jun Chen, Longshuai Xiao, Chao Weng, Wei Xue, Lei Xie. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

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/.

Recent progress in text-to-speech (TTS) has achieved impressive naturalness and flexibility, especially with the development of large language model (LLM)-based approaches. However, existing autoregressive (AR) structures and large-scale models, such as Llasa, still face significant challenges in inference latency and streaming synthesis. To deal with the limitations, we introduce Llasa+, an accelerated and streaming TTS model built on Llasa. Specifically, to accelerate the generation process, we introduce two plug-and-play Multi-Token Prediction (MTP) modules following the frozen backbone. These modules allow the model to predict multiple tokens in one AR step. Additionally, to mitigate potential error propagation caused by inaccurate MTP, we design a novel verification algorithm that leverages the frozen backbone to validate the generated tokens, thus allowing Llasa+ to achieve speedup without sacrificing generation quality. Furthermore, we design a causal decoder that enables streaming speech reconstruction from tokens. Extensive experiments show that Llasa+ achieves a $1.48 \times$ speedup without sacrificing generation quality, despite being trained only on LibriTTS. Moreover, the MTP-and-verification framework can be applied to accelerate any LLM-based model. All codes and models are publicly available at https://github.com/ASLP-lab/LLaSA_Plus.

Photorealistic 3D human modeling is essential for various applications and has seen tremendous progress. However, existing methods for monocular full-body reconstruction, typically relying on front and/or predicted back view, still struggle with satisfactory performance due to the ill-posed nature of the problem and sophisticated self-occlusions. In this paper, we propose PSHuman, a novel framework that explicitly reconstructs human meshes utilizing priors from the multiview diffusion model. It is found that directly applying multiview diffusion on single-view human images leads to severe geometric distortions, especially on generated faces. To address it, we propose a cross-scale diffusion that models the joint probability distribution of global full-body shape and local facial characteristics, enabling identity-preserved novel-view generation without geometric distortion. Moreover, to enhance cross-view body shape consistency of varied human poses, we condition the generative model on parametric models (SMPL-X), which provide body priors and prevent unnatural views inconsistent with human anatomy. Leveraging the generated multiview normal and color images, we present SMPLX-initialized explicit human carving to recover realistic textured human meshes efficiently. Extensive experiments on CAPE and THuman2.1 demonstrate PSHuman’s superiority in geometry details, texture fidelity, and generalization capability.

In this work, we systematically study music generation conditioned solely on the video. First, we present a large-scale dataset by collecting 360K video-music pairs, including various genres such as movie trailers, advertisements, and documentaries. Furthermore, we propose VidMuse, a simple framework for generating music aligned with video inputs. VidMuse stands out by producing high-fidelity music that is both acoustically and semantically aligned with the video. By incorporating local and global visual cues, VidMuse enables the creation of coherent music tracks that consistently match the video content through Long-Short-Term modeling. Through extensive experiments, VidMuse outperforms existing models in terms of audio quality, diversity, and audio-visual alignment. The code and datasets are available at https://vidmuse.github.io/

Recently, the pretrain-finetune paradigm has become a cornerstone in various deep learning areas.While in general the pre-trained model would promote both effectiveness and efficiency of downstream tasks fine-tuning, studies have shown that not all knowledge acquired during pre-training is beneficial.Some of the knowledge may actually bring detrimental effects to the fine-tuning tasks, which is also known as negative transfer.To address this problem, graceful forgetting has emerged as a promising approach.The core principle of graceful forgetting is to enhance the learning plasticity of the target task by selectively discarding irrelevant knowledge.However, this approach remains underexplored in the context of generative language models, and it is often challenging to migrate existing forgetting algorithms to these models due to architecture incompatibility.To bridge this gap, in this paper we propose a novel framework, Learning With Forgetting (LWF), to achieve graceful forgetting in generative language models.With Fisher Information Matrix weighting the intended parameter updates, LWF computes forgetting confidence to evaluate selfgenerated knowledge regarding the forgetting task, and consequently, knowledge with high confidence is periodically unlearned during fine-tuning.Our experiments demonstrate that, although thoroughly uncovering the mechanisms of knowledge interaction remains challenging in pre-trained language models, applying graceful forgetting can contribute to enhanced fine-tuning performance.

Despite the significant progress in controllable music generation and editing, challenges remain in the quality and length of generated music due to the use of Mel-spectrogram representations and UNet-based model structures. To address these limitations, we propose a novel approach using a Diffusion Transformer (DiT) augmented with an additional control branch using ControlNet. This allows for long-form and variable-length music generation and editing controlled by text and melody prompts. For more precise and fine-grained melody control, we introduce a novel top-k constant-Q Transform representation as the melody prompt, reducing ambiguity compared to previous representations (e.g., chroma), particularly for music with multiple tracks or a wide range of pitch values. To effectively balance the control signals from text and melody prompts, we adopt a curriculum learning strategy that progressively masks the melody prompt, resulting in a more stable training process. Experiments have been performed on text-to-music generation and music-style transfer tasks using open-source instrumental recording data. The results demonstrate that by extending StableAudio, a pre-trained text-controlled DiT model, our approach enables superior melody-controlled editing while retaining good text-to-music generation performance. These results outperform a strong MusicGen baseline in terms of both text-based generation and melody preservation for editing. Audio examples can be found at https://stable-audio-control.github.io.

Low-Rank Adaptation (LoRA) is a widely used method for efficiently fine-tuning large models by introducing lowrank matrices into weight updates. However, existing LoRA techniques fail to account for activation information, such as outliers, which significantly impact model performance. This omission leads to suboptimal adaptation and slower convergence. To address this limitation, we present Activation-Informed Low-Rank Adaptation (AIRA), a novel approach that integrates activation information into initialization, training, and rank assignment to enhance model performance. Specifically, AIRA introduces: (1) Outlierweighted SVD decomposition to reduce approximation errors in low-rank weight initialization, (2) Outlier-driven dynamic rank assignment using offline optimization for better layer-wise adaptation, and (3) Activation-informed training to amplify updates on significant weights. This cascaded activation-informed paradigm enables faster convergence and fewer fine-tuned parameters while maintaining high performance. Extensive experiments on multiple large models demonstrate that AIRA outperforms state-of-the-art LoRA variants, achieving superior performance-efficiency trade-offs in vision-language instruction tuning, few-shot learning, and image generation. Codes are available at https://github.com/lliai/LoRA-Zoo.

In this paper, we present the first structural binarization method for LLM compression to less than 1-bit precision. Although LLMs have achieved remarkable performance, their memory-bound nature during the inference stage hinders the adoption of resource-constrained devices. Reducing weights to 1-bit precision through binarization substantially enhances computational efficiency. We observe that randomly flipping some weights in binarized LLMs does not significantly degrade the model's performance, suggesting the potential for further compression. To exploit this, our STBLLM employs an N:M sparsity technique to achieve structural binarization of the weights. Specifically, we introduce a novel Standardized Importance (SI) metric, which considers weight magnitude and input feature norm to more accurately assess weight significance. Then, we propose a layer-wise approach, allowing different layers of the LLM to be sparsified with varying N:M ratios, thereby balancing compression and accuracy. Furthermore, we implement a fine-grained grouping strategy for less important weights, applying distinct quantization schemes to sparse, intermediate, and dense regions. Finally, we design a specialized CUDA kernel to support structural binarization. We conduct extensive experiments on LLaMA, OPT, and Mistral family. STBLLM achieves a perplexity of 11.07 at 0.55 bits per weight, outperforming the BiLLM by 3×. The results demonstrate that our approach performs better than other compressed binarization LLM methods while significantly reducing memory requirements. Code is released at https://github.com/pprp/STBLLM.

In this paper, we explore the application of Large Language Models (LLMs) to the pre-training of music. While the prevalent use of MIDI in music modeling is well-established, our findings suggest that LLMs are inherently more compatible with ABC Notation, which aligns more closely with their design and strengths, thereby enhancing the model's performance in musical composition. To address the challenges associated with misaligned measures from different tracks during generation, we propose the development of a $\underline{S}$ynchronized $\underline{M}$ulti-$\underline{T}$rack ABC Notation ($\textbf{SMT-ABC Notation}$), which aims to preserve coherence across multiple musical tracks. Our contributions include a series of models capable of handling up to 8192 tokens, covering 90\% of the symbolic music data in our training set. Furthermore, we explore the implications of the $\underline{S}$ymbolic $\underline{M}$usic $\underline{S}$caling Law ($\textbf{SMS Law}$) on model performance. The results indicate a promising research direction in music generation, offering extensive resources for further research through our open-source contributions.

Recently, diffusion models have achieved great success in mono-channel audio generation. However, when it comes to stereo audio generation, the soundscapes often have a complex scene of multiple objects and directions. Controlling stereo audio with spatial contexts remains challenging due to high data costs and unstable generative models. To the best of our knowledge, this work represents the first attempt to address these issues. We first construct a large-scale, simulation-based, and GPT-assisted dataset, BEWO-1M, with abundant soundscapes and descriptions even including moving and multiple sources. Beyond text modality, we have also acquired a set of images and rationally paired stereo audios through retrieval to advance multimodal generation. Existing audio generation models tend to generate rather random and indistinct spatial audio. To provide accurate guidance for Latent Diffusion Models, we introduce the SpatialSonic model utilizing spatial-aware encoders and azimuth state matrices to reveal reasonable spatial guidance. By leveraging spatial guidance, our model not only achieves the objective of generating immersive and controllable spatial audio from text but also extends to other modalities as the pioneer attempt. Finally, under fair settings, we conduct subjective and objective evaluations on simulated and real-world data to compare our approach with prevailing methods. The results demonstrate the effectiveness of our method, highlighting its capability to generate spatial audio that adheres to physical rules.

Video generation models have made significant progress in simulating future states, showcasing their potential as world simulators in embodied scenarios. However, existing models often lack robust understanding, limiting their ability to perform multi-step predictions or handle Out-of-Distribution (OOD) scenarios. To address this challenge, we propose the Reflection of Generation (RoG), a set of intermediate reasoning strategies designed to enhance video prediction. It leverages the complementary strengths of pre-trained vision-language and video generation models, enabling them to function as a world model in embodied scenarios. To support RoG, we introduce Embodied Video Anticipation Benchmark(EVA-Bench), a comprehensive benchmark that evaluates embodied world models across diverse tasks and scenarios, utilizing both in-domain and OOD datasets. Building on this foundation, we devise a world model, Embodied Video Anticipator (EVA), that follows a multistage training paradigm to generate high-fidelity video frames and apply an autoregressive strategy to enable adaptive generalization for longer video sequences. Extensive experiments demonstrate the efficacy of EVA in various downstream tasks like video generation and robotics, thereby paving the way for large-scale pre-trained models in real-world video prediction applications. The video demos are available at https://sites.google.com/view/icml-eva.

Mixture-of-Experts (MoE) architectures in large language models (LLMs) achieve exceptional performance, but face prohibitive storage and memory requirements. To address these challenges, we present $D^2$-MoE, a new delta decompression compressor for reducing the parameters of MoE LLMs. Based on observations of expert diversity, we decompose their weights into a shared base weight and unique delta weights. Specifically, our method first merges each expert's weight into the base weight using the Fisher information matrix to capture shared components. Then, we compress delta weights through Singular Value Decomposition (SVD) by exploiting their low-rank properties. Finally, we introduce a semi-dynamical structured pruning strategy for the base weights, combining static and dynamic redundancy analysis to achieve further parameter reduction while maintaining input adaptivity. In this way, our $D^2$-MoE successfully compacts MoE LLMs to high compression ratios without additional training. Extensive experiments highlight the superiority of our approach, with over 13\% performance gains than other compressors on Mixtral|Phi-3.5|DeepSeek|Qwen2 MoE LLMs at 40$\sim$60\% compression rates. Codes are available in https://github.com/lliai/D2MoE.

Mixture of Experts (MoE) architecture improves Large Language Models (LLMs) with better scaling, but its higher parameter counts and memory demands create challenges for deployment. In this paper, we present MoE-SVD, a new decomposition-based compression framework tailored for MoE LLMs without any extra training. By harnessing the power of Singular Value Decomposition (SVD), MoE-SVD addresses the critical issues of decomposition collapse and matrix redundancy in MoE architectures. Specifically, we first decompose experts into compact low-rank matrices, resulting in accelerated inference and memory optimization. In particular, we propose selective decomposition strategy by measuring sensitivity metrics based on weight singular values and activation statistics to automatically identify decomposable expert layers. Then, we share a single V-matrix across all experts and employ a top-k selection for U-matrices. This low-rank matrix sharing and trimming scheme allows for significant parameter reduction while preserving diversity among experts. Comprehensive experiments on Mixtral, Phi-3.5, DeepSeek, and Qwen2 MoE LLMs show MoE-SVD outperforms other compression methods, achieving a 60% compression ratio and 1.5$\times$ faster inference with minimal performance loss.

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.

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.

Facial personalization faces challenges to maintain identity fidelity without disrupting the foundation model's prompt consistency. The mainstream personalization models employ identity embedding to integrate identity information within the attention mechanisms. However, our preliminary findings reveal that identity embeddings compromise the effectiveness of other tokens in the prompt, thereby limiting high prompt consistency and attribute-level controllability. Moreover, by deactivating identity embedding, personalization models still demonstrate the underlying foundation models' ability to control facial attributes precisely. It suggests that such foundation models' knowledge can be leveraged to cure the ill-aligned prompt consistency of personalization models. Building upon these insights, we propose FreeCure, a framework that improves the prompt consistency of personalization models with their latent foundation models' knowledge. First, by setting a dual inference paradigm with/without identity embedding, we identify attributes (e.g., hair, accessories, etc.) for enhancements. Second, we introduce a novel foundation-aware self-attention module, coupled with an inversion-based process to bring well-aligned attribute information to the personalization process. Our approach is training-free, and can effectively enhance a wide array of facial attributes; and it can be seamlessly integrated into existing popular personalization models based on both Stable Diffusion and FLUX. FreeCure has consistently shown significant improvements in prompt consistency across these facial personalization models while maintaining the integrity of their original identity fidelity.

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.

We introduce MMAR, a new benchmark designed to evaluate the deep reasoning capabilities of Audio-Language Models (ALMs) across massive multi-disciplinary tasks. MMAR comprises 1,000 meticulously curated audio-question-answer triplets, collected from real-world internet videos and refined through iterative error corrections and quality checks to ensure high quality. Unlike existing benchmarks that are limited to specific domains of sound, music, or speech, MMAR extends them to a broad spectrum of real-world audio scenarios, including mixed-modality combinations of sound, music, and speech. Each question in MMAR is hierarchically categorized across four reasoning layers: Signal, Perception, Semantic, and Cultural, with additional sub-categories within each layer to reflect task diversity and complexity. To further foster research in this area, we annotate every question with a Chain-of-Thought (CoT) rationale to promote future advancements in audio reasoning. Each item in the benchmark demands multi-step deep reasoning beyond surface-level understanding. Moreover, a part of the questions requires graduate-level perceptual and domain-specific knowledge, elevating the benchmark's difficulty and depth. We evaluate MMAR using a broad set of models, including Large Audio-Language Models (LALMs), Large Audio Reasoning Models (LARMs), Omni Language Models (OLMs), Large Language Models (LLMs), and Large Reasoning Models (LRMs), with audio caption inputs. The performance of these models on MMAR highlights the benchmark's challenging nature, and our analysis further reveals critical limitations of understanding and reasoning capabilities among current models. These findings underscore the urgent need for greater research attention in audio-language reasoning, including both data and algorithm innovation. We hope MMAR will serve as a catalyst for future advances in this important but little-explored area.

Multi-person motion capture over sparse angular observations is a challenging problem under interference from both self- and mutual-occlusions. Existing works produce accurate 2D joint detection, however, when these are triangulated and lifted into 3D, available solutions all struggle in selecting the most accurate candidates and associating them to the correct joint type and target identity. As such, in order to fully utilize all accurate 2D joint location information, we propose to independently triangulate between all same-typed 2D joints from all camera views regardless of their target ID, forming the Joint Cloud. Joint Cloud consist of both valid joints lifted from the same joint type and target ID, as well as falsely constructed ones that are from different 2D sources. These redundant and inaccurate candidates are processed over the proposed Joint Cloud Selection and Aggregation Transformer (JCSAT) involving three cascaded encoders which deeply explore the trajectile, skeletal structural, and view-dependent correlations among all 3D point candidates in the cross-embedding space. An Optimal Token Attention Path (OTAP) module is proposed which subsequently selects and aggregates informative features from these redundant observations for the final prediction of human motion. To demonstrate the effectiveness of JCSAT, we build and publish a new multi-person motion capture dataset BUMocap-X with complex interactions and severe occlusions. Comprehensive experiments over the newly presented as well as benchmark datasets validate the effectiveness of the proposed framework, which outperforms all existing state-of-the-art methods, especially under challenging occlusion scenarios.

The superior performances of pre-trained foundation models in various visual tasks underscore their potential to enhance the 2D models' open-vocabulary ability. Existing methods explore analogous applications in the 3D space. However, most of them only center around knowledge extraction from singular foundation models, which limits the open-vocabulary ability of 3D models. We hypothesize that leveraging complementary pre-trained knowledge from various foundation models can improve knowledge transfer from 2D pre-trained visual language models to the 3D space. In this work, we propose FM-OV3D, a method of Foundation Model-based Cross-modal Knowledge Blending for Open-Vocabulary 3D Detection, which improves the open-vocabulary localization and recognition abilities of 3D model by blending knowledge from multiple pre-trained foundation models, achieving true open-vocabulary without facing constraints from original 3D datasets. Specifically, to learn the open-vocabulary 3D localization ability, we adopt the open-vocabulary localization knowledge of the Grounded-Segment-Anything model. For open-vocabulary 3D recognition ability, We leverage the knowledge of generative foundation models, including GPT-3 and Stable Diffusion models, and cross-modal discriminative models like CLIP. The experimental results on two popular benchmarks for open-vocabulary 3D object detection show that our model efficiently learns knowledge from multiple foundation models to enhance the open-vocabulary ability of the 3D model and successfully achieves state-of-the-art performance in open-vocabulary 3D object detection tasks. Code is released at https://github.com/dmzhang0425/FM-OV3D.git.

Generating vivid and emotional 3D co-speech gestures is crucial for virtual avatar animation in human-machine interaction applications. While the existing methods enable generating the gestures to follow a single emotion label, they overlook that long gesture sequence modeling with emotion transition is more practical in real scenes. In addition, the lack of large-scale available datasets with emotional transition speech and corresponding 3D human gestures also limits the addressing of this task. To fulfill this goal, we first incorporate the ChatGPT-4 and an audio inpainting approach to construct the high-fidelity emotion transition human speeches. Considering obtaining the realistic 3D pose annotations corresponding to the dynamically inpainted emotion transition audio is extremely difficult, we propose a novel weakly supervised training strategy to encourage authority gesture transitions. Specifically, to enhance the coordination of transition gestures w. r. t. different emotional ones, we model the temporal association representation between two different emotional gesture sequences as style guidance and infuse it into the transition generation. We further devise an emotion mixture mechanism that provides weak supervision based on a learnable mixed emotion label for transition gestures. Last, we present a keyframe sampler to supply effective initial posture cues in long sequences, enabling us to generate diverse gestures. Extensive experiments demonstrate that our method outperforms the state-of-the-art models constructed by adapting single emotion-conditioned counterparts on our newly defined emotion transition task and datasets. Our code and dataset will be released on the project page: https://xingqunqi-lab.github.io/Emo-Transition-Gesture/.

Text evaluation has historically posed significant challenges, often demanding substantial labor and time cost. With the emergence of large language models (LLMs), researchers have explored LLMs' potential as alternatives for human evaluation. While these single-agent-based approaches show promise, experimental results suggest that further advancements are needed to bridge the gap between their current effectiveness and human-level evaluation quality. Recognizing that best practices of human evaluation processes often involve multiple human annotators collaborating in the evaluation, we resort to a multi-agent debate framework, moving beyond single-agent prompting strategies. In this paper, we construct a multi-agent referee team called $\textbf{ChatEval}$ to autonomously discuss and evaluate the quality of different texts. Our experiments on two benchmarks illustrate that ChatEval delivers superior accuracy and correlation in alignment with human assessment. Furthermore, we find that the diverse role prompts (different personas) are essential in the multi-agent debate process; that is, utilizing the same role description in the prompts can lead to a degradation in performance. Our qualitative analysis also shows that ChatEval transcends mere textual scoring, offering a human-mimicking evaluation process for reliable assessments.

Since real-world machine systems are running in non-stationary environments, Continual Test-Time Adaptation (CTTA) task is proposed to adapt the pre-trained model to continually changing target domains. Recently, existing methods mainly focus on model-based adaptation, which aims to leverage a self-training manner to extract the target domain knowledge. However, pseudo labels can be noisy and the updated model parameters are unreliable under dynamic data distributions, leading to error accumulation and catastrophic forgetting in the continual adaptation process. To tackle these challenges and maintain the model plasticity, we design a Visual Domain Adapter (ViDA) for CTTA, explicitly handling both domain-specific and domain-shared knowledge. Specifically, we first comprehensively explore the different domain representations of the adapters with trainable high-rank or low-rank embedding spaces. Then we inject ViDAs into the pre-trained model, which leverages high-rank and low-rank features to adapt the current domain distribution and maintain the continual domain-shared knowledge, respectively. To exploit the low-rank and high-rank ViDAs more effectively, we further propose a Homeostatic Knowledge Allotment (HKA) strategy, which adaptively combines different knowledge from each ViDA. Extensive experiments conducted on four widely used benchmarks demonstrate that our proposed method achieves state-of-the-art performance in both classification and segmentation CTTA tasks. Note that, our method can be regarded as a novel transfer paradigm for large-scale models, delivering promising results in adaptation to continually changing distributions.

In this paper, we present DetKDS, the first framework that searches for optimal detection distillation policies. Manual design of detection distillers becomes challenging and time-consuming due to significant disparities in distillation behaviors between detectors with different backbones, paradigms, and label assignments. To tackle these challenges, we leverage search algorithms to discover optimal distillers for homogeneous and heterogeneous student-teacher pairs. Firstly, our search space encompasses global features, foreground-background features, instance features, logits response, and localization response as inputs. Then, we construct omni-directional cascaded transformations and obtain the distiller by selecting the advanced distance function and common weight value options. Finally, we present a divide-and-conquer evolutionary algorithm to handle the explosion of the search space. In this strategy, we first evolve the best distiller formulations of individual knowledge inputs and then optimize the combined weights of these multiple distillation losses. DetKDS automates the distillation process without requiring expert design or additional tuning, effectively reducing the teacher-student gap in various scenarios. Based on the analysis of our search results, we provide valuable guidance that contributes to detection distillation designs. Comprehensive experiments on different detectors demonstrate that DetKDS outperforms state-of-the-art methods in detection and instance segmentation tasks. For instance, DetKDS achieves significant gains than baseline detectors: $+3.7$, $+4.1$, $+4.0$, $+3.7$, and $+3.5$ AP on RetinaNet, Faster-RCNN, FCOS, RepPoints, and GFL, respectively. Code at: https://github.com/lliai/DetKDS.

Recent progress in large-scale zero-shot speech synthesis has been significantly advanced by language models and diffusion models. However, the generation process of both methods is slow and computationally intensive. Efficient speech synthesis using a lower computing budget to achieve quality on par with previous work remains a significant challenge. In this paper, we present FlashSpeech, a large-scale zero-shot speech synthesis system with approximately 5% of the inference time compared with previous work. FlashSpeech is built on the latent consistency model and applies a novel adversarial consistency training approach that can train from scratch without the need for a pre-trained diffusion model as the teacher. Furthermore, a new prosody generator module enhances the diversity of prosody, making the rhythm of the speech sound more natural. The generation processes of FlashSpeech can be achieved efficiently with one or two sampling steps while maintaining high audio quality and high similarity to the audio prompt for zero-shot speech generation. Our experimental results demonstrate the superior performance of FlashSpeech. Notably, FlashSpeech can be about 20 times faster than other zero-shot speech synthesis systems while maintaining comparable performance in terms of voice quality and similarity. Furthermore, FlashSpeech demonstrates its versatility by efficiently performing tasks like voice conversion, speech editing, and diverse speech sampling. Audio samples can be found in https://flashspeech.github.io/

In this paper, we present DSA, the first automated framework for discovering sparsity allocation schemes for layer-wise pruning in Large Language Models (LLMs). LLMs have become increasingly powerful, but their large parameter counts make them computationally expensive. Existing pruning methods for compressing LLMs primarily focus on evaluating redundancies and removing element-wise weights. However, these methods fail to allocate adaptive layer-wise sparsities, leading to performance degradation in challenging tasks. We observe that per-layer importance statistics can serve as allocation indications, but their effectiveness depends on the allocation function between layers. To address this issue, we develop an expression discovery framework to explore potential allocation strategies. Our allocation functions involve two steps: reducing element-wise metrics to per-layer importance scores, and modelling layer importance to sparsity ratios. To search for the most effective allocation function, we construct a search space consisting of pre-process, reduction, transform, and post-process operations. We leverage an evolutionary algorithm to perform crossover and mutation on superior candidates within the population, guided by performance evaluation. Finally, we seamlessly integrate our discovered functions into various uniform methods, resulting in significant performance improvements. We conduct extensive experiments on multiple challenging tasks such as arithmetic, knowledge reasoning, and multimodal benchmarks spanning GSM8K, MMLU, SQA, and VQA, demonstrating that our DSA method achieves significant performance gains on the LLaMA-1|2|3, Mistral, and OPT models. Notably, the LLaMA-1|2|3 model pruned by our DSA reaches 4.73\%|6.18\%|10.65\% gain over the state-of-the-art techniques (e.g., Wanda and SparseGPT).

In this paper, we introduce **Era3D**, a novel multiview diffusion method that generates high-resolution multiview images from a single-view image. Despite significant advancements in multiview generation, existing methods still suffer from camera prior mismatch, inefficacy, and low resolution, resulting in poor-quality multiview images. Specifically, these methods assume that the input images should comply with a predefined camera type, e.g. a perspective camera with a fixed focal length, leading to distorted shapes when the assumption fails. Moreover, the full-image or dense multiview attention they employ leads to a dramatic explosion of computational complexity as image resolution increases, resulting in prohibitively expensive training costs. To bridge the gap between assumption and reality, Era3D first proposes a diffusion-based camera prediction module to estimate the focal length and elevation of the input image, which allows our method to generate images without shape distortions. Furthermore, a simple but efficient attention layer, named row-wise attention, is used to enforce epipolar priors in the multiview diffusion, facilitating efficient cross-view information fusion. Consequently, compared with state-of-the-art methods, Era3D generates high-quality multiview images with up to a 512×512 resolution while reducing computation complexity of multiview attention by 12x times. Comprehensive experiments demonstrate the superior generation power of Era3D- it can reconstruct high-quality and detailed 3D meshes from diverse single-view input images, significantly outperforming baseline multiview diffusion methods.

Catastrophic forgetting poses a significant challenge in continual learning (CL). In the context of Natural Language Processing, generative-based rehearsal CL methods have made progress in avoiding expensive retraining. However, generating pseudo samples that accurately capture the task-specific distribution remains a daunting task. In this paper, we propose Dirichlet Continual Learning (DCL), a novel generative-based rehearsal strategy designed specifically for CL. Different from the conventional use of Gaussian latent variable in Conditional Variational Autoencoder, DCL employs the flexibility of the Dirichlet distribution to model the latent variable. This allows DCL to effectively capture sentence-level features from previous tasks and guide the generation of pseudo samples. Additionally, we introduce Jensen-Shannon Knowledge Distillation, a robust logit-based knowledge distillation method that enhances knowledge transfer during pseudo-sample generation. Our extensive experiments show that DCL outperforms state-of-the-art methods in two typical tasks of task-oriented dialogue systems, demonstrating its efficacy.

Cross-modal Retrieval (CMR) is formulated for the scenarios where the queries and retrieval results are of different modalities. Existing Cross-modal Retrieval (CMR) studies mainly focus on the common contextualized information between text transcripts and images, and the synchronized event information in audio-visual recordings. Unlike all previous works, in this article, we investigate the geometric correspondence between images and speech recordings captured in the same space and formulate a novel CMR task, called Spatial Image-Acoustic Retrieval (SIAR). To this end, we first design a novel speech encoder that consists of convolution neural networks and transformer layers, to learn space-aware speech representations. Then, to eliminate the cross-modal inherent discrepancy, we propose the Contrastive Speech Image Retrieval (CSIR) method which uses supervised contrastive learning to attract the same-space cross-modal features while repelling the ones from different spaces. Finally, image and speech features are directly compared and we predict the SIAR result with the maximum similarity. Extensive experiments demonstrate that our proposed speech encoder can recognize space from human speeches with superior performance over the other prevailing networks. It also sets our penultimate goal of speech-to-speech retrieval. Furthermore, our CSIR proposal can successfully perform bi-directional SIAR between spatial images and reverberant speeches with promising results. Code and data will be available.

The significant development of artificial neural network architectures has facilitated the increasing adoption of automated music composition models over the past few years. However, most existing systems feature algorithmic generative structures based on hard code and predefined rules, generally excluding interactive or improvised behaviors. We propose a motion based music system, MoMusic, as a AI real time music generation system. MoMusic features a partially randomized harmonic sequencing model based on a probabilistic analysis of tonal chord progressions, mathematically abstracted through musical set theory. This model is presented against a dual dimension grid that produces resulting sounds through a posture recognition mechanism. A camera captures the users' fingers' movement and trajectories, creating coherent, partially improvised harmonic progressions. MoMusic integrates several timbrical registers, from traditional classical instruments such as the piano to a new ''human voice instrument'' created using a voice conversion technique. Our research demonstrates MoMusic's interactiveness, ability to inspire musicians, and ability to generate coherent musical material with various timbrical registers. MoMusic's capabilities could be easily expanded to incorporate different forms of posture controlled timbrical transformation, rhythmic transformation, dynamic transformation, or even digital sound processing techniques.

Existing noise-robust and reverberant-robust localization algorithms fail to localize the target speaker when interfering speakers are present. In this paper, we address the problem of localizing only the target speaker in multi-speaker scenarios and propose a target speaker localization algorithm, called GCC-speaker. Specifically, we modify the weighting of the generalized cross-correlation with phase transform (GCC-PHAT) algorithm and propose an optimal speaker-dependent weighting based on a novel localization-related loss function and data-driven training. The speaker-dependent weighting is responsible for guiding the GCC algorithm to obtain the optimal target speaker localization results. As for the loss function, we constrain the estimated GCC angular spectrum and the estimated direction of arrival (DOA) to be close to their ground truth values, respectively. The experimental results show the superiority of GCC-speaker compared to the existing target speaker localization algorithms for different signal-to-interference ratios, reverberation times and array geometries.

Developing digital sound synthesizers is crucial to the music industry as it provides a low-cost way to produce high-quality sounds with rich timbres. Existing traditional synthesizers often require substantial expertise to determine the overall framework of a synthesizer and the parameters of submodules. Since expert knowledge is hard to acquire, it hinders the flexibility to quickly design and tune digital synthesizers for diverse sounds. In this paper, we propose ``NAS-FM'', which adopts neural architecture search (NAS) to build a differentiable frequency modulation (FM) synthesizer. Tunable synthesizers with interpretable controls can be developed automatically from sounds without any prior expert knowledge and manual operating costs. In detail, we train a supernet with a specifically designed search space, including predicting the envelopes of carriers and modulators with different frequency ratios. An evolutionary search algorithm with adaptive oscillator size is then developed to find the optimal relationship between oscillators and the frequency ratio of FM. Extensive experiments on recordings of different instrument sounds show that our algorithm can build a synthesizer fully automatically, achieving better results than handcrafted synthesizers. Audio samples are available at https://nas-fm.github.io/

Denoising diffusion probabilistic models (DDPMs) have shown promising performance for speech synthesis. However, a large number of iterative steps are required to achieve high sample quality, which restricts the inference speed. Maintaining sample quality while increasing sampling speed has become a challenging task. In this paper, we propose a Consistency Model-based Speech synthesis method, CoMoSpeech, which achieve speech synthesis through a single diffusion sampling step while achieving high audio quality. The consistency constraint is applied to distill a consistency model from a well-designed diffusion-based teacher model, which ultimately yields superior performances in the distilled CoMoSpeech. Our experiments show that by generating audio recordings by a single sampling step, the CoMoSpeech achieves an inference speed more than 150 times faster than real-time on a single NVIDIA A100 GPU, which is comparable to FastSpeech2, making diffusion-sampling based speech synthesis truly practical. Meanwhile, objective and subjective evaluations on text-to-speech and singing voice synthesis show that the proposed teacher models yield the best audio quality, and the one-step sampling based CoMoSpeech achieves the best inference speed with better or comparable audio quality to other conventional multi-step diffusion model baselines. Audio samples and codes are available at https://comospeech.github. https://comospeech.github.io/.

In the era of extensive intersection between art and Artificial Intelligence (AI), such as image generation and fiction co-creation, AI for music remains relatively nascent, particularly in music understanding. This is evident in the limited work on deep music representations, the scarcity of large-scale datasets, and the absence of a universal and community-driven benchmark. To address this issue, we introduce the Music Audio Representation Benchmark for universaL Evaluation, termed MARBLE. It aims to provide a benchmark for various Music Information Retrieval (MIR) tasks by defining a comprehensive taxonomy with four hierarchy levels, including acoustic, performance, score, and high-level description. We then establish a unified protocol based on 18 tasks on 12 public-available datasets, providing a fair and standard assessment of representations of all open-sourced pre-trained models developed on music recordings as baselines. Besides, MARBLE offers an easy-to-use, extendable, and reproducible suite for the community, with a clear statement on copyright issues on datasets. Results suggest recently proposed large-scale pre-trained musical language models perform the best in most tasks, with room for further improvement. The leaderboard and toolkit repository are published to promote future music AI research.

Generalized Cross Correlation (GCC) is the most popular localization technique over the past decades and can be extended with the beamforming method e.g. Steered Response Power (SRP) when multiple microphone pairs exist. Considering the promising results of Deep Learning (DL) strategies over classical approaches, in this work, instead of directly using Generalized Cross Correlation (GCC), SRP is derived with the DL-learnt ideal correlation functions for each pair of a microphone array. To deploy visual information, we explore the Conditional Variational Auto-Encoder (CVAE) framework in which the audio generative process is conditioned on the visual features encoded by face detections. The vision-derived auxiliary correlation function eventually contributes to the back-end beamformer for improved localization performance. To the best of our knowledge, this is the first deep-generative audiovisual method for speaker localization. Experimental results demonstrate our superior performance over other competitive methods, especially when the speech signal is corrupted by noise.

Conventional learning-based speech enhancement methods usually utilize existing building blocks to design the deep neural networks (DNNs), while how to effectively integrate the statistical signal processing based schemes, which are expert-knowledge driven and could ameliorate the over-fitting problem, into the network design remains an open issue. In this paper, we extend the conventional Kalman filtering (KF) and propose a supervised-learning based neural Kalman filter (NKF) for speech enhancement. Similar to KF, the proposed method first obtains a prediction from the speech evolution model and then integrates the short-term instantaneous observation by linear weighting, and the weights are calculated by comparing between the speech prediction residual error and the environmental noise level. An end-to-end network is designed to convert the speech linear prediction model in KF to non-linear, and to compact all other conventional linear filtering operations. Different with other DNN based methods, the proposed method provides a specialized network design inspired from the conventional signal processing, the backpropagation can be directly applied on the linear filtering operations integrated from KF. We conduct experiments in different noisy conditions, and the results demonstrate that the proposed method outperforms the baseline methods which are based on either signal processing or DNNs.

Recently we presented a modulation-domain multichannel Kalman filtering (MKF) algorithm for speech enhancement, which jointly exploits the inter-frame modulation-domain temporal evolution of speech and the inter-channel spatial correlation to estimate the clean speech signal. The goal of speech enhancement is to suppress noise while keeping the speech undistorted, and a key problem is to achieve the best trade-off between speech distortion and noise reduction. In this paper, we extend the MKF by presenting a modulation-domain parametric MKF (PMKF) which includes a parameter that enables flexible control of the speech enhancement behaviour in each time-frequency (TF) bin. Based on the decomposition of the MKF cost function, a new cost function for PMKF is proposed, which uses the controlling parameter to weight the noise reduction and speech distortion terms. An optimal PMKF gain is derived using a minimum mean squared error (MMSE) criterion. We analyse the performance of the proposed MKF, and show its relationship to the speech distortion weighted multichannel Wiener filter (SDW-MWF). To evaluate the impact of the controlling parameter on speech enhancement performance, we further propose PMKF speech enhancement systems in which the controlling parameter is adaptively chosen in each TF bin. Experiments on a publicly available head-related impulse response (HRIR) database in different noisy and reverberant conditions demonstrate the effectiveness of the proposed method.

This paper presents the development of our systems for the Interspeech 2020 Far-Field Speaker Verification Challenge (FFSVC). Our focus is the task 2 of the challenge, which is to perform far-field text-independent speaker verification using a single microphone array. The FFSVC training set provided by the challenge is augmented by pre-processing the far-field data with both beamforming, voice channel switching, and a combination of weighted prediction error (WPE) and beamforming. Two open-access corpora, CHData in Mandarin and VoxCeleb2 in English, are augmented using multiple methods and mixed with the augmented FFSVC data to form the final training data. Four different model structures are used to model speaker characteristics: ResNet, extended time-delay neural network (ETDNN), Transformer, and factorized TDNN (FTDNN), whose output values are pooled across time using the self-attentive structure, the statistic pooling structure, and the GVLAD structure. The final results are derived by fusing the adaptively normalized scores of the four systems with a two-stage fusion method, which achieves a minimum of the detection cost function (minDCF) of 0.3407 and an equal error rate (EER) of 2.67% on the development set of the challenge.

The performance of sound event localization and detection (SELD) degrades in source-overlapping cases since features of different sources collapse with each other, and the network tends to fail to learn to separate these features effectively. In this paper, by leveraging the conventional microphone array signal processing to generate comprehensive representations for SELD, we propose a new SELD method based on multiple direction of arrival (DOA) beamforming and multi-task learning. By using multiple beamformers to extract the signals from different DOAs, the sound field is more diversely described, and specialised representations of target source and noises can be obtained. With labelled training data, the steering vector is estimated based on the cross-power spectra (CPS) and the signal presence probability (SPP), which eliminates the need of knowing the array geometry. We design two networks for sound event localization (SED) and sound source localization (SSL) and use a multi-task learning scheme for SED, in which the SSL-related task act as a regularization. Experimental results using the database of DCASE2019 SELD task show that the proposed method achieves the state-of-art performance.

Sound source localization (SSL) is challenging in presence of reverberation since the cross-correlation between the direct-path signals in different microphones, which indicates the spatial information of the sound source, is interfered by the reverberation signal components. A novel algorithm is proposed in this paper to estimate the cross-correlation of the direct-path speech signals, such that the robustness of SSL to reverberation can be improved. The proposed method follows a similar scheme to the multichannel linear prediction (MCLP), which is commonly used for speech dereverberation, while avoids the explicit estimation of the direct-path signal of each channel. This is achieved by revealing the relationship between the direct-path signal cross-correlation (DPCC) and the MCLP coefficient vector, and finally deriving the DPCC by using only the multichannel reverberant signals. It is also shown that the pre-whitening operation, which is widely used for SSL, can be inherently integrated into the estimated DPCC. An adaptive method is further derived to facilitate online frame-level SSL. The proposed method can be easily applied to conventional cross-correlation based SSL methods by using the DPCC rather than the full cross-correlation. Experiments conducted in various reverberant conditions demonstrate the effectiveness of the proposed method.

The noise covariance matrix computed between the signals from a microphone array is used in the design of spatial filters and beamformers with applications in noise suppression and dereverberation. This paper specifically addresses the problem of estimating the covariance matrix associated with a noise field when the array is rotating during desired source activity, as is common in head-mounted arrays. We propose a parametric model that leads to an analytical expression for the microphone signal covariance as a function of the array orientation and array manifold. An algorithm for estimating the model parameters during noise-only segments is proposed and the performance shown to be improved, rather than degraded, by array rotation. The stored model parameters can then be used to update the covariance matrix to account for the effects of any array rotation that occurs when the desired source is active. The proposed method is evaluated in terms of the Frobenius norm of the error in the estimated covariance matrix and of the noise reduction performance of a minimum variance distortionless response beamformer. In simulation experiments the proposed method achieves 18 dB lower error in the estimated noise covariance matrix than a conventional recursive averaging approach and results in noise reduction which is within 0.05 dB of an oracle beamformer using the ground truth noise covariance matrix.

The use of spatial information in multichannel speech enhancement methods is well established but information associated with the temporal evolution of speech is less commonly exploited. Speech signals can be modelled using an autoregressive process in the time-frequency modulation domain, and Kalman filtering based speech enhancement algorithms have been developed for single-channel processing. In this paper, a multichannel Kalman filter (MKF) for speech enhancement is derived that jointly considers the multichannel spatial information and the temporal correlations of speech. We model the temporal evolution of speech in the modulation domain and, by incorporating the spatial information, an optimal MKF gain is derived in the short-time Fourier transform domain. We also show that the proposed MKF becomes a conventional multichannel Wiener filter if the temporal information is discarded. Experiments using the signals generated from a public head-related impulse response database demonstrate the effectiveness of the proposed method in comparison to other techniques.

Compared with single-channel speech enhancement methods, multichannel methods can utilize spatial information to design optimal filters. Although some filters adaptively consider second-order signal statistics, the temporal evolution of the speech spectrum is usually neglected. By using linear prediction (LP) to model the inter-frame temporal evolution of speech, single-channel Kalman filtering (KF) based methods have been developed for speech enhancement. In this paper, we derive a multichannel KF (MKF) that jointly uses both interchannel spatial correlation and interframe temporal correlation for speech enhancement. We perform LP in the modulation domain, and by incorporating the spatial information, derive an optimal MKF gain in the short-time Fourier transform domain. We show that the proposed MKF reduces to the conventional multichannel Wiener filter if the LP information is discarded. Furthermore, we show that, under an appropriate assumption, the MKF is equivalent to a concatenation of the minimum variance distortion response beamformer and a single-channel modulation-domain KF and therefore present an alternative implementation of the MKF. Experiments conducted on a public head-related impulse response database demonstrate the effectiveness of the proposed method.

In room acoustics, under-modelled multichannel blind system identification (BSI) aims to estimate the early part of the room impulse responses (RIRs), and it can be widely used in applications such as speaker localization, room geometry identification and beamforming based speech dereverberation. In this paper we extend our recent study on under-modelled BSI from the time domain to the frequency domain, such that the RIRs can be updated frame-wise and the efficiency of Fast Fourier Transform (FFT) is exploited to reduce the computational complexity. Analogous to the cross-correlation based criterion in the time domain, a frequency-domain cross power spectrum based criterion is proposed. As the early RIRs are usually sparse, the RIRs are estimated by jointly maximizing the cross power spectrum based criterion in the frequency domain and minimizing the l1-norm sparsity measure in the time domain. A two-stage LMS updating algorithm is derived to achieve joint optimization of these two targets. The experimental results in different under-modelled scenarios demonstrate the effectiveness of the proposed method.