This paper introduces briefly the history and growth of the Detection and Classification of Acoustic Scenes and Events (DCASE) challenge, workshop, research area and research community. Created in 2013 as a data evaluation challenge, DCASE has become a major research topic in the Audio and Acoustic Signal Processing area. Its success comes from a combination of factors: the challenge offers a large variety of tasks that are renewed each year; and the workshop offers a channel for dissemination of related work, engaging a young and dynamic community. At the same time, DCASE faces its own challenges, growing and expanding to different areas. One of the core principles of DCASE is open science and reproducibility: publicly available datasets, baseline systems, technical reports and workshop publications. While the DCASE challenge and workshop are independent of IEEE SPS, the challenge receives annual endorsement from the AASP TC, and the DCASE community contributes significantly to the ICASSP flagship conference and the success of SPS in many of its activities.

Language-queried audio source separation (LASS) focuses on separating sounds using textual descriptions of the desired sources. Current methods mainly use discriminative approaches, such as time-frequency masking, to separate target sounds and minimize interference from other sources. However, these models face challenges when separating overlapping sound-tracks, which may lead to artifacts such as spectral holes or incomplete separation. Rectified flow matching (RFM), a generative model that establishes linear relations between the distribution of data and noise, offers superior theoretical properties and simplicity, but has not yet been explored in sound separation. In this work, we introduce FlowSep, a new generative model based on RFM for LASS tasks. FlowSep learns linear flow trajectories from noise to target source features within the variational autoencoder (VAE) latent space. During inference, the RFM-generated latent features are reconstructed into a mel-spectrogram via the pre-trained VAE decoder, followed by a pre-trained vocoder to synthesize the waveform. Trained on 1, 680 hours of audio data, FlowSep outperforms the state-of-the-art models across multiple benchmarks, as evaluated with subjective and objective metrics. Additionally, our results show that FlowSep surpasses a diffusion-based LASS model in both separation quality and inference efficiency, highlighting its strong potential for audio source separation tasks. Code, pre-trained models and demos can be found at: https://audio-agi.github.io/FlowSep_demo/.

Generative models have shown significant achievements in audio generation tasks. However, existing models struggle with complex and detailed prompts, leading to potential performance degradation. We hypothesize that this problem stems from the simplicity and scarcity of the training data. This work aims to create a large-scale audio dataset with rich captions for improving audio generation models. We first develop an automated pipeline to generate detailed captions by transforming predicted visual captions, audio captions, and tagging labels into comprehensive descriptions using a Large Language Model (LLM). The resulting dataset, Sound-VECaps, comprises 1.66M high-quality audio-caption pairs with enriched details including audio event orders, occurred places and environment information. We then demonstrate that training the text-to-audio generation models with Sound-VECaps significantly improves the performance on complex prompts. Furthermore, we conduct ablation studies of the models on several downstream audio-language tasks, showing the potential of Sound-VECaps in advancing audio-text representation learning.Dataset and demos are available at https://yyua8222.github.io/Sound-VECaps-demo/.

We introduce Music Source Restoration (MSR), a novel task addressing the gap between idealized source separation and real-world music production. Current Music Source Separation (MSS) approaches assume mixtures are simple sums of sources, ignoring signal degradations employed during music production like equalization, compression, and reverb. MSR models mixtures as degraded sums of individually degraded sources, with the goal of recovering original, undegraded signals. Due to the lack of data for MSR, we present RawStems, a dataset annotation of 578 songs with unprocessed source signals organized into 8 primary and 17 secondary instrument groups, totaling 354.13 hours. To the best of our knowledge, RawStems is the first dataset that contains unprocessed music stems with hierarchical categories. We consider spectral filtering, dynamic range compression, harmonic distortion, reverb and lossy codec as possible degradations, and establish U-Former as a baseline method, demonstrating the feasibility of MSR on our dataset. We release the RawStems dataset annotations, degradation simulation pipeline, training code and pre-trained models to be publicly available.

The audio spectrogram is a time-frequency representation that has been widely used for audio classification. One of the key attributes of the audio spectrogram is the temporal resolution, which depends on the hop size used in the Short-Time Fourier Transform (STFT). Previous works generally assume the hop size should be a constant value (e.g., 10 ms). However, a fixed temporal resolution is not always optimal for different types of sound. The temporal resolution affects not only classification accuracy but also computational cost. This paper proposes a novel method, DiffRes, that enables differentiable temporal resolution modeling for audio classification. Given a spectrogram calculated with a fixed hop size, DiffRes merges non-essential time frames while preserving important frames. DiffRes acts as a "drop-in" module between an audio spectrogram and a classifier and can be jointly optimized with the classification task. We evaluate DiffRes on five audio classification tasks, using mel-spectrograms as the acoustic features, followed by off-the-shelf classifier backbones. Compared with previous methods using the fixed temporal resolution, the DiffRes-based method can achieve the equivalent or better classification accuracy with at least 25% computational cost reduction. We further show that DiffRes can improve classification accuracy by increasing the temporal resolution of input acoustic features, without adding to the computational cost.

Audio super-resolution is a fundamental task that predicts high-frequency components for low-resolution audio, enhancing audio quality in digital applications. Previous methods have limitations such as the limited scope of audio types (e.g., music, speech) and specific bandwidth settings they can handle (e.g., 4 kHz to 8 kHz). In this paper, we introduce a diffusion-based generative model, AudioSR, that is capable of performing robust audio super-resolution on versatile audio types, including sound effects, music, and speech. Specifically, AudioSR can upsample any input audio signal within the bandwidth range of 2 kHz to 16 kHz to a high-resolution audio signal at 24 kHz bandwidth with a sampling rate of 48 kHz. Extensive objective evaluation on various audio super-resolution benchmarks demonstrates the strong result achieved by the proposed model. In addition, our subjective evaluation shows that AudioSR can act as a plug-and-play module to enhance the generation quality of a wide range of audio generative models, including AudioLDM, Fastspeech2, and MusicGen. Our code and demo are available at https://audioldm.github.io/audiosr.

Despite recent progress in text-to-audio (TTA) generation, we show that the state-of-the-art models, such as AudioLDM, trained on datasets with an imbalanced class distribution, such as AudioCaps, are biased in their generation performance. Specifically, they excel in generating common audio classes while underperforming in the rare ones, thus degrading the overall generation performance. We refer to this problem as long-tailed text-to-audio generation. To address this issue, we propose a simple retrieval-augmented approach for TTA models. Specifically, given an input text prompt, we first leverage a Contrastive Language Audio Pretraining (CLAP) model to retrieve relevant text-audio pairs. The features of the retrieved audio-text data are then used as additional conditions to guide the learning of TTA models. We enhance AudioLDM with our proposed approach and denote the resulting augmented system as Re-AudioLDM. On the AudioCaps dataset, Re-AudioLDM achieves a state-of-the-art Frechet Audio Distance (FAD) of 1.37, outperforming the existing approaches by a large margin. Furthermore, we show that Re-AudioLDM can generate realistic audio for complex scenes, rare audio classes, and even unseen audio types, indicating its potential in TTA tasks.

Transformers, which were originally developed for natural language processing, have recently generated significant interest in the computer vision and audio communities due to their flexibility in learning long-range relationships. Constrained by the data hungry nature of transformers and the limited amount of labelled data, most transformer-based models for audio tasks are finetuned from ImageNet pretrained models, despite the huge gap between the domain of natural images and audio. This has motivated the research in self-supervised pretraining of audio transformers, which reduces the dependency on large amounts of labeled data and focuses on extracting concise representations of audio spectrograms. In this paper, we proposeLocal-GlobalAudioSpectrogram vIsionTransformer, namely ASiT, a novel self-supervised learning framework that captures local and global contextual information by employing group masked model learning and self-distillation. We evaluate our pretrained models on both audio and speech classification tasks, including audio event classification, keyword spotting, and speaker identification. We further conduct comprehensive ablation studies, including evaluations of different pretraining strategies. The proposed ASiT framework significantly boosts the performance on all tasks and sets a new state-of-the-art performance in five audio and speech classification tasks, outperforming recent methods, including the approaches that use additional datasets for pretraining.

Environment shifts and conflicts present significant challenges for learning-based sound event localization and detection (SELD) methods. SELD systems, when trained in particular acoustic settings, often show restricted generalization capabilities for diverse acoustic environments. Furthermore, obtaining annotated samples for spatial sound events is notably costly. Deploying a SELD system in a new environment requires extensive time for re-training and fine-tuning. To overcome these challenges, we propose environment-adaptive Meta-SELD, designed for efficient adaptation to new environments using minimal data. Our method specifically utilizes computationally synthesized spatial data and employs Model-Agnostic Meta-Learning (MAML) on a pre-trained, environment-independent model. The method then utilizes fast adaptation to unseen real-world environments using limited samples from the respective environments. Inspired by the Learning-to-Forget approach, we introduce the concept of selective memory as a strategy for resolving conflicts across environments. This approach involves selectively memorizing target-environment-relevant information and adapting to the new environments through the selective attenuation of model parameters. In addition, we introduce environment representations to characterize different acoustic settings, enhancing the adaptability of our attenuation approach to various environments. We evaluate our proposed method on the development set of the Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23) dataset and computationally synthesized scenes. Experimental results demonstrate the superior performance of the proposed method compared to conventional supervised learning methods, particularly in localization.

Although audio generation shares commonalities across different types of audio, such as speech, music, and sound effects, designing models for each type requires careful consideration of specific objectives and biases that can significantly differ from those of other types. To bring us closer to a unified perspective of audio generation, this paper proposes a holistic framework that utilizes the same learning method for speech, music, and sound effect generation. Our framework utilizes a general representation of audio, called “language of audio” (LOA). Any audio can be translated into LOA based on AudioMAE, a self-supervised pre-trained representation learning model. In the generation process, we translate other modalities into LOA by using a GPT-2 model, and we perform self-supervised audio generation learning with a latent diffusion model conditioned on the LOA of audio in our training set. The proposed framework naturally brings advantages such as reusable self-supervised pretrained latent diffusion models. Experiments on the major benchmarks of text-to-audio, text-to-music, and text-to-speech with three AudioLDM 2 variants demonstrate competitive performance of the AudioLDM 2 variants framework against previous approaches. Our code, pretrained model, and demo are available athttps://audioldm.github.io/audioldm2.

Automated audio captioning is a cross-modal translation task for describing the content of audio clips with natural language sentences. This task has attracted increasing attention and substantial progress has been made in recent years. Captions generated by existing models are generally faithful to the content of audio clips, however, these machine-generated captions are often deterministic (e.g., generating a fixed caption for a given audio clip), simple (e.g., using common words and simple grammar), and generic (e.g., generating the same caption for similar audio clips). When people are asked to describe the content of an audio clip, different people tend to focus on different sound events and describe an audio clip diversely from various aspects using distinct words and grammar. We believe that an audio captioning system should have the ability to generate diverse captions, either for a fixed audio clip, or across similar audio clips. To this end, we propose an adversarial training framework based on a conditional generative adversarial network (C-GAN) to improve diversity of audio captioning systems. A caption generator and two hybrid discriminators compete and are learned jointly, where the caption generator can be any standard encoder-decoder captioning model used to generate captions, and the hybrid discriminators assess the generated captions from different criteria, such as their naturalness and semantics. We conduct experiments on the Clotho dataset. The results show that our proposed model can generate captions with better diversity as compared to state-of-the-art methods.

The advancement of audio-language (AL) multimodal learning tasks has been significant in recent years, yet the limited size of existing audio-language datasets poses challenges for researchers due to the costly and time-consuming collection process. To address this data scarcity issue, we introduceWavCaps, the first large-scale weakly-labelled audio captioning dataset, comprising approximately 400 k audio clips with paired captions. We sourced audio clips and their raw descriptions from web sources and a sound event detection dataset. However, the online-harvested raw descriptions are highly noisy and unsuitable for direct use in tasks such as automated audio captioning. To overcome this issue, we propose a three-stage processing pipeline for filtering noisy data and generating high-quality captions, where ChatGPT, a large language model, is leveraged to filter and transform raw descriptions automatically. We conduct a comprehensive analysis of the characteristics of WavCaps dataset and evaluate it on multiple downstream audio-language multimodal learning tasks. The systems trained on WavCaps outperform previous state-of-the-art (SOTA) models by a significant margin. Our aspiration is for the WavCaps dataset we have proposed to facilitate research in audio-language multimodal learning and demonstrate the potential of utilizing large language models (LLMs) to enhance academic research.

Acoustic Scene Classification (ASC) is a task that classifies a scene according to environmental acoustic signals. Audios collected from different cities and devices often exhibit biases in feature distributions, which may negatively impact ASC performance. Taking the city and device of the audio collection as two types of data domain, this paper attempts to disentangle the audio features of each domain to remove the related feature biases. A dual-alignment framework is proposed to generalize the ASC system on new devices or cities, by aligning boundaries across domains and decision boundaries within each domain. During the alignment, the maximum classifier discrepancy and gradient reversed layer are used for the feature disentanglement of scene, city and device, while four candidate domain classifiers are proposed to explore the optimal solution of feature disentanglement. To evaluate the dual-alignment framework, three experiments of biased ASC tasks are designed: 1) cross-city ASC in new cities; 2) cross-device ASC in new devices; 3) cross-city-device ASC in new cities and new devices. Results demonstrate the superiority of the proposed framework, showcasing performance improvements of 0.9%, 19.8%, and 10.7% on classification accuracy, respectively. The effectiveness of the proposed feature disentanglement approach is further evaluated in both biased and unbiased ASC problems, and the results demonstrate that better-disentangled audio features can lead to a more robust ASC system across different devices and cities. This paper advocates for the integration of feature disentanglement in ASC systems to achieve more reliable performance.

Recently, there has been increasing interest in building efficient audio neural networks for on-device scenarios. Most existing approaches are designed to reduce the size of audio neural networks using methods such as model pruning. In this work, we show that instead of reducing model size using complex methods, eliminating the temporal redundancy in the input audio features (e.g., mel-spectrogram) could be an effective approach for efficient audio classification. To do so, we proposed a family of simple pooling front-ends (SimPFs) which use simple non-parametric pooling operations to reduce the redundant information within the mel-spectrogram. We perform extensive experiments on four audio classification tasks to evaluate the performance of SimPFs. Experimental results show that SimPFs can achieve a reduction in more than half of the number of floating point operations (FLOPs) for off-the-shelf audio neural networks, with negligible degradation or even some improvements in audio classification performance.

Convolution neural networks (CNNs) have shown great success in various applications. However, the computational complexity and memory storage of CNNs is a bottleneck for their deployment on resource-constrained devices. Recent efforts towards reducing the computation cost and the memory overhead of CNNs involve similarity-based passive filter pruning methods. Similarity-based passive filter pruning methods compute a pairwise similarity matrix for the filters and eliminate a few similar filters to obtain a small pruned CNN. However, the computational complexity of computing the pairwise similarity matrix is high, particularly when a convolutional layer has many filters. To reduce the computational complexity in obtaining the pairwise similarity matrix, we propose to use an efficient method where the complete pairwise similarity matrix is approximated from only a few of its columns by using a Nyström approximation method. The proposed efficient similarity-based passive filter pruning method is 3 times faster and gives same accuracy at the same reduction in computations for CNNs compared to that of the similarity-based pruning method that computes a complete pairwise similarity matrix. Apart from this, the proposed efficient similarity-based pruning method performs similarly or better than the existing norm-based pruning methods. The efficacy of the proposed pruning method is evaluated on CNNs such as DCASE 2021 Task 1A baseline network and a VGGish network designed for acoustic scene classification.

Text-to-audio (TTA) systems have recently gained attention for their ability to synthesize general audio based on text descriptions. However, previous studies in TTA have limited generation quality with high computational costs. In this study, we propose AudioLDM, a TTA system that is built on a latent space to learn continuous audio representations from contrastive language-audio pretraining (CLAP) embeddings. The pretrained CLAP models enable us to train LDMs with audio embeddings while providing text embeddings as the condition during sampling. By learning the latent representations of audio signals without modelling the cross-modal relationship, AudioLDM improves both generation quality and computational efficiency. Trained on AudioCaps with a single GPU, AudioLDM achieves state-of-the-art TTA performance compared to other open-sourced systems, measured by both objective and subjective metrics. AudioLDM is also the first TTA system that enables various text-guided audio manipulations (e.g., style transfer) in a zero-shot fashion. Our implementation and demos are available at https://audioldm.github.io.

Contrastive language-audio pretraining (CLAP) has become a new paradigm to learn audio concepts with audio-text pairs. CLAP models have shown unprecedented performance as zero-shot classifiers on downstream tasks. To further adapt CLAP with domain-specific knowledge, a popular method is to finetune its audio encoder with available labelled examples. However, this is challenging in low-shot scenarios, as the amount of annotations is limited compared to the model size. In this work, we introduce a Training-efficient (Treff) adapter to rapidly learn with a small set of examples while maintaining the capacity for zero-shot classification. First, we propose a cross-attention linear model (CALM) to map a set of labelled examples and test audio to test labels. Second, we find initialising CALM as a cosine measurement improves our Treff adapter even without training. The Treff adapter outperforms metric-based methods in few-shot settings and yields competitive results to fully-supervised methods.

Audio captioning aims to generate text descriptions of audio clips.In the real world, many objects produce similar sounds.How to accurately recognize ambiguous sounds is a major challenge for audio captioning.In this work, inspired by inherent human multimodal perception, we propose visuallyaware audio captioning, which makes use of visual information to help the description of ambiguous sounding objects.Specifically, we introduce an off-the-shelf visual encoder to extract video features and incorporate the visual features into an audio captioning system.Furthermore, to better exploit complementary audio-visual contexts, we propose an audio-visual attention mechanism that adaptively integrates audio and visual context and removes the redundant information in the latent space.Experimental results on AudioCaps, the largest audio captioning dataset, show that our proposed method achieves state-of-theart results on machine translation metrics.

This study defines a new evaluation metric for audio tagging tasks to alleviate the limitation of the mean average precision (mAP) metric.The mAP metric treats different kinds of sound as independent classes without considering their relations.The proposed metric, ontology-aware mean average precision (OmAP), addresses the weaknesses of mAP by utilizing additional ontology during evaluation.Specifically, we reweight the false positive events in the model prediction based on the AudioSet ontology graph distance to the target classes.The OmAP also provides insights into model performance by evaluating different coarse-grained levels in the ontology graph.We conduct a human assessment and show that OmAP is more consistent with human perception than mAP.We also propose an ontology-based loss function (OBCE) that reweights binary cross entropy (BCE) loss based on the ontology distance.Our experiment shows that OBCE can improve both mAP and OmAP metrics on the AudioSet tagging task.

Automated audio captioning (AAC) which generates textual descriptions of audio content.Existing AAC models achieve good results but only use the high-dimensional representation of the encoder.There is always insufficient information learning of high-dimensional methods owing to high-dimensional representations having a large amount of information.In this paper, a new encoder-decoder model called the Lowand High-Dimensional Feature Fusion (LHDFF) is proposed.LHDFF uses a new PANNs encoder called Residual PANNs (RPANNs) to fuse low-and high-dimensional features.Lowdimensional features contain limited information about specific audio scenes.The fusion of low-and high-dimensional features can improve model performance by repeatedly emphasizing specific audio scene information.To fully exploit the fused features, LHDFF uses a dual transformer decoder structure to generate captions in parallel.Experimental results show that LHDFF outperforms existing audio captioning models.

Polyphonic sound event localization and detection (SELD) aims at detecting types of sound events with corresponding temporal activities and spatial locations. In this paper, a trackwise ensemble event independent network with a novel data augmentation method is proposed. The proposed model is based on our previous proposed Event-Independent Network V2 and is extended by conformer blocks and dense blocks. The track-wise ensemble model with track-wise output format is proposed to solve an ensemble model problem for track-wise output format that track permutation may occur among different models. The data augmentation approach contains several data augmentation chains, which are composed of random combinations of several data augmentation operations. The method also utilizes log-mel spectrograms, intensity vectors, and Spatial Cues-Augmented Log-Spectrogram (SALSA) for different models. We evaluate our proposed method in the Task of the L3DAS22 challenge and obtain the top ranking solution with a location-dependent F-score to be 0.699. Source code is released1.

Audio captioning aims at generating natural language descriptions for audio clips automatically. Existing audio captioning models have shown promising improvement in recent years. However, these models are mostly trained via maximum likelihood estimation (MLE), which tends to make captions generic, simple and deterministic. As different people may describe an audio clip from different aspects using distinct words and grammars, we argue that an audio captioning system should have the ability to generate diverse captions for a fixed audio clip and across similar audio clips. To address this problem, we propose an adversarial training framework for audio captioning based on a conditional generative adversarial network (C-GAN), which aims at improving the naturalness and diversity of generated captions. Unlike processing data of continuous values in a classical GAN, a sentence is composed of discrete tokens and the discrete sampling process is non-differentiable. To address this issue, policy gradient, a reinforcement learning technique, is used to back-propagate the reward to the generator. The results show that our proposed model can generate more diverse captions, as compared to state-of-the-art methods.

In this paper, we introduce the task of language-queried audio source separation (LASS), which aims to separate a target source from an audio mixture based on a natural language query of the target source (e.g., "a man tells a joke followed by people laughing"). A unique challenge in LASS is associated with the complexity of natural language description and its relation with the audio sources. To address this issue, we proposed LASS-Net, an end-to-end neural network that is learned to jointly process acoustic and linguistic information, and separate the target source that is consistent with the language query from an audio mixture. We evaluate the performance of our proposed system with a dataset created from the AudioCaps dataset. Experimental results show that LASS-Net achieves considerable improvements over baseline methods. Furthermore, we observe that LASS-Net achieves promising generalization results when using diverse human-annotated descriptions as queries, indicating its potential use in real-world scenarios. The separated audio samples and source code are available at https://liuxubo717.github.io/LASS-demopage.

Audio-text retrieval aims at retrieving a target audio clip or caption from a pool of candidates given a query in another modality. Solving such cross-modal retrieval task is challenging because it not only requires learning robust feature representations for both modalities, but also requires capturing the fine-grained alignment between these two modalities. Existing cross-modal retrieval models are mostly optimized by metric learning objectives as both of them attempt to map data to an embedding space, where similar data are close together and dissimilar data are far apart. Unlike other cross-modal retrieval tasks such as image-text and video-text retrievals, audio-text retrieval is still an unexplored task. In this work, we aim to study the impact of different metric learning objectives on the audio-text retrieval task. We present an extensive evaluation of popular metric learning objectives on the AudioCaps and Clotho datasets. We demonstrate that NT-Xent loss adapted from self-supervised learning shows stable performance across different datasets and training settings, and outperforms the popular triplet-based losses. Our code is available at https://github.com/XinhaoMei/audio-text_retrieval.

We present a method to develop low-complexity convolutional neural networks (CNNs) for acoustic scene classification (ASC).The large size and high computational complexity of typical CNNs is a bottleneck for their deployment on resourceconstrained devices.We propose a passive filter pruning framework, where a few convolutional filters from the CNNs are eliminated to yield compressed CNNs.Our hypothesis is that similar filters produce similar responses and give redundant information allowing such filters to be eliminated from the network.To identify similar filters, a cosine distance based greedy algorithm is proposed.A fine-tuning process is then performed to regain much of the performance lost due to filter elimination.To perform efficient fine-tuning, we analyze how the performance varies as the number of fine-tuning training examples changes.An experimental evaluation of the proposed framework is performed on the publicly available DCASE 2021 Task 1A baseline network trained for ASC.The proposed method is simple, reduces computations per inference by 27%, with 25% fewer parameters, with less than 1% drop in accuracy.

Polyphonic sound event localization and detection (SELD), which jointly performs sound event detection (SED) and direction-of-arrival (DoA) estimation, detects the type and occurrence time of sound events as well as their corresponding DoA angles simultaneously. We study the SELD task from a multi-task learning perspective. Two open problems are addressed in this paper. Firstly, to detect overlapping sound events of the same type but with different DoAs, we propose to use a trackwise output format and solve the accompanying track permutation problem with permutation-invariant training. Multi-head self-attention is further used to separate tracks. Secondly, a previous finding is that, by using hard parameter-sharing, SELD suffers from a performance loss compared with learning the subtasks separately. This is solved by a soft parameter-sharing scheme. We term the proposed method as Event Independent Network V2 (EINV2), which is an improved version of our previously-proposed method and an end-to-end network for SELD. We show that our proposed EINV2 for joint SED and DoA estimation outperforms previous methods by a large margin, and has comparable performance to state-of-the-art ensemble models.

In real rooms, recorded speech usually contains reverberation, which degrades the quality and intelligibility of the speech. It has proven effective to use neural networks to estimate complex ideal ratio masks (cIRMs) using mean square error (MSE) loss for speech dereverberation. However, in some cases, when using MSE loss to estimate complex-valued masks, phase may have a disproportionate effect compared to magnitude. We propose a new weighted magnitude-phase loss function, which is divided into a magnitude component and a phase component, to train a neural network to estimate complex ideal ratio masks. A weight parameter is introduced to adjust the relative contribution of magnitude and phase to the overall loss. We find that our proposed loss function outperforms the regular MSE loss function for speech dereverberation.

Energy disaggregation, a.k.a. Non-Intrusive Load Monitoring, aims to separate the energy consumption of individual appliances from the readings of a mains power meter measuring the total energy consumption of, e.g., a whole house. Energy consumption of individual appliances can be useful in many applications, e.g., providing appliance-level feedback to the end users to help them understand their energy consumption and ultimately save energy. Recently, with the availability of large-scale energy consumption datasets, various neural network models such as convolutional neural networks and recurrent neural networks have been investigated to solve the energy disaggregation problem. Neural network models can learn complex patterns from large amounts of data and have been shown to outperform the traditional machine learning methods such as variants of hidden Markov models. However, current neural network methods for energy disaggregation are either computational expensive or are not capable of handling long-term dependencies. In this article, we investigate the application of the recently developed WaveNet models for the task of energy disaggregation. Based on a real-world energy dataset collected from 20 households over 2 years, we show that WaveNet models outperforms the state-of-the-art deep learning methods proposed in the literature for energy disaggregation in terms of both error measures and computational cost. On the basis of energy disaggregation, we then investigate the performance of two deep-learning based frameworks for the task of on/off detection which aims at estimating whether an appliance is in operation or not. The first framework obtains the on/off states of an appliance by binarising the predictions of a regression model trained for energy disaggregation, while the second framework obtains the on/off states of an appliance by directly training a binary classifier with binarised energy readings of the appliance serving as the target values. Based on the same dataset, we show that for the task of on/off detection the second framework, i.e., directly training a binary classifier, achieves better performance in terms of F1 score.

Acoustic Scene Classification (ASC) aims to classify the environment in which the audio signals are recorded. Recently, Convolutional Neural Networks (CNNs) have been successfully applied to ASC. However, the data distributions of the audio signals recorded with multiple devices are different. There has been little research on the training of robust neural networks on acoustic scene datasets recorded with multiple devices, and on explaining the operation of the internal layers of the neural networks. In this article, we focus on training and explaining device-robust CNNs on multi-device acoustic scene data. We propose conditional atrous CNNs with attention for multi-device ASC. Our proposed system contains an ASC branch and a device classification branch, both modelled by CNNs. We visualise and analyse the intermediate layers of the atrous CNNs. A time-frequency attention mechanism is employed to analyse the contribution of each time-frequency bin of the feature maps in the CNNs. On the Detection and Classification of Acoustic Scenes and Events (DCASE) 2018 ASC dataset, recorded with three devices, our proposed model performs significantly better than CNNs trained on single-device data.

In supervised machine learning, the assumption that training data is labelled correctly is not always satisfied. In this paper, we investigate an instance of labelling error for classification tasks in which the dataset is corrupted with out-of-distribution (OOD) instances: data that does not belong to any of the target classes, but is labelled as such. We show that detecting and relabelling certain OOD instances, rather than discarding them, can have a positive effect on learning. The proposed method uses an auxiliary classifier, trained on data that is known to be in-distribution, for detection and relabelling. The amount of data required for this is shown to be small. Experiments are carried out on the FSDnoisy18k audio dataset, where OOD instances are very prevalent. The proposed method is shown to improve the performance of convolutional neural networks by a significant margin. Comparisons with other noise-robust techniques are similarly encouraging.

Source separation is the task of separating an audio recording into individual sound sources. Source separation is fundamental for computational auditory scene analysis. Previous work on source separation has focused on separating particular sound classes such as speech and music. Much previous work requires mixtures and clean source pairs for training. In this work, we propose a source separation framework trained with weakly labelled data. Weakly labelled data only contains the tags of an audio clip, without the occurrence time of sound events. We first train a sound event detection system with AudioSet. The trained sound event detection system is used to detect segments that are most likely to contain a target sound event. Then a regression is learnt from a mixture of two randomly selected segments to a target segment conditioned on the audio tagging prediction of the target segment. Our proposed system can separate 527 kinds of sound classes from AudioSet within a single system. A U-Net is adopted for the separation system and achieves an average SDR of 5.67 dB over 527 sound classes in AudioSet.

Audio pattern recognition is an important research topic in the machine learning area, and includes several tasks such as audio tagging, acoustic scene classification, music classification, speech emotion classification and sound event detection. Recently, neural networks have been applied to tackle audio pattern recognition problems. However, previous systems are built on specific datasets with limited durations. Recently, in computer vision and natural language processing, systems pretrained on large-scale datasets have generalized well to several tasks. However, there is limited research on pretraining systems on large-scale datasets for audio pattern recognition. In this paper, we propose pretrained audio neural networks (PANNs) trained on the large-scale AudioSet dataset. These PANNs are transferred to other audio related tasks. We investigate the performance and computational complexity of PANNs modeled by a variety of convolutional neural networks. We propose an architecture called Wavegram-Logmel-CNN using both log-mel spectrogram and waveform as input feature. Our best PANN system achieves a state-of-the-art mean average precision (mAP) of 0.439 on AudioSet tagging, outperforming the best previous system of 0.392. We transfer PANNs to six audio pattern recognition tasks, and demonstrate state-of-the-art performance in several of those tasks. We have released the source code and pretrained models of PANNs:https://github.com/qiuqiangkong/audioset_tagging_cnn.

Sound event detection (SED) is a task to detect sound events in an audio recording. One challenge of the SED task is that many datasets such as the Detection and Classification of Acoustic Scenes and Events (DCASE) datasets are weakly labelled. That is, there are only audio tags for each audio clip without the onset and offset times of sound events. We compare segment-wise and clip-wise training for SED that is lacking in previous works. We propose a convolutional neural network transformer (CNN-Transfomer) for audio tagging and SED, and show that CNN-Transformer performs similarly to a convolutional recurrent neural network (CRNN). Another challenge of SED is that thresholds are required for detecting sound events. Previous works set thresholds empirically, and are not an optimal approaches. To solve this problem, we propose an automatic threshold optimization method. The first stage is to optimize the system with respect to metrics that do not depend on thresholds, such as mean average precision (mAP). The second stage is to optimize the thresholds with respect to metrics that depends on those thresholds. Our proposed automatic threshold optimization system achieves a state-of-the-art audio tagging F1 of 0.646, outperforming that without threshold optimization of 0.629, and a sound event detection F1 of 0.584, outperforming that without threshold optimization of 0.564.

Sound event detection (SED) methods typically rely on either strongly labelled data or weakly labelled data. As an alternative, sequentially labelled data (SLD) was proposed. In SLD, the events and the order of events in audio clips are known, without knowing the occurrence time of events. This paper proposes a connectionist temporal classification (CTC) based SED system that uses SLD instead of strongly labelled data, with a novel unsupervised clustering stage. Experiments on 41 classes of sound events show that the proposed two-stage method trained on SLD achieves performance comparable to the previous state-of-the-art SED system trained on strongly labelled data, and is far better than another state-of-the-art SED system trained on weakly labelled data, which indicates the effectiveness of the proposed two-stage method trained on SLD without any onset/offset time of sound events.

Acoustic scene generation (ASG) is a task to generate waveforms for acoustic scenes. ASG can be used to generate audio scenes for movies and computer games. Recently, neural networks such as SampleRNN have been used for speech and music generation. However, ASG is more challenging due to its wide variety. In addition, evaluating a generative model is also difficult. In this paper, we propose to use a conditional SampleRNN model to generate acoustic scenes conditioned on the input classes. We also propose objective criteria to evaluate the quality and diversity of the generated samples based on classification accuracy. The experiments on the DCASE 2016 Task 1 acoustic scene data show that with the generated audio samples, a classification accuracy of 65.5% can be achieved compared to samples generated by a random model of 6.7% and samples from real recording of 83.1%. The performance of a classifier trained only on generated samples achieves an accuracy of 51.3%, as opposed to an accuracy of 6.7% with samples generated by a random model.

Humans are able to identify a large number of environmental sounds and categorise them according to high-level semantic categories, e.g. urban sounds or music. They are also capable of generalising from past experience to new sounds when applying these categories. In this paper we report on the creation of a data set that is structured according to the top-level of a taxonomy derived from human judgements and the design of an associated machine learning challenge, in which strong generalisation abilities are required to be successful. We introduce a baseline classification system, a deep convolutional network, which showed strong performance with an average accuracy on the evaluation data of 80.8%. The result is discussed in the light of two alternative explanations: An unlikely accidental category bias in the sound recordings or a more plausible true acoustic grounding of the high-level categories.

Acoustic Event Detection (AED) is an important task of machine listening which, in recent years, has been addressed using common machine learning methods like Non-negative Matrix Factorization (NMF) or deep learning. However, most of these approaches do not take into consideration the way that human auditory system detects salient sounds. In this work, we propose a method for AED using weakly labeled data that combines a Non-negative Matrix Factorization model with a salience model based on predictive coding in the form of Kalman filters. We show that models of auditory perception, particularly auditory salience, can be successfully incorporated into existing AED methods and improve their performance on rare event detection. We evaluate the method on the Task2 of DCASE2017 Challenge.

The goal of Acoustic Scene Classification (ASC) is to recognise the environment in which an audio waveform has been recorded. Recently, deep neural networks have been applied to ASC and have achieved state-of-the-art performance. However, few works have investigated how to visualise and understand what a neural network has learnt from acoustic scenes. Previous work applied local pooling after each convolutional layer, therefore reduced the size of the feature maps. In this paper, we suggest that local pooling is not necessary, but the size of the receptive field is important. We apply atrous Convolutional Neural Networks (CNNs) with global attention pooling as the classification model. The internal feature maps of the attention model can be visualised and explained. On the Detection and Classification of Acoustic Scenes and Events (DCASE) 2018 dataset, our proposed method achieves an accuracy of 72.7 %, significantly outperforming the CNNs without dilation at 60.4 %. Furthermore, our results demonstrate that the learnt feature maps contain rich information on acoustic scenes in the time-frequency domain.

Single-channel signal separation and deconvolution aims to separate and deconvolve individual sources from a single-channel mixture. Single-channel signal separation and deconvolution is a challenging problem in which no prior knowledge of the mixing filters is available. Both individual sources and mixing filters need to be estimated. In addition, a mixture may contain non-stationary noise which is unseen in the training set. We propose a synthesizing-decomposition (S-D) approach to solve the single-channel separation and deconvolution problem. In synthesizing, a generative model for sources is built using a generative adversarial network (GAN). In decomposition, both mixing filters and sources are optimized to minimize the reconstruction error of the mixture. The proposed S-D approach achieves a peak-to-noise-ratio (PSNR) of 18.9 dB and 15.4 dB in image inpainting and completion, outperforming a baseline convolutional neural network PSNR of 15.3 dB and 12.2 dB, respectively and achieves a PSNR of 13.2 dB in source separation together with deconvolution, outperforming a convolutive non-negative matrix factorization (NMF) baseline of 10.1 dB.

Sound event detection (SED) aims to detect when and recognize what sound events happen in an audio clip. Many supervised SED algorithms rely on strongly labelled data that contains the onset and offset annotations of sound events. However, many audio tagging datasets are weakly labelled, that is, only the presence of the sound events is known, without knowing their onset and offset annotations. In this paper, we propose a time–frequency (T–F) segmentation framework trained on weakly labelled data to tackle the sound event detection and separation problem. In training, a segmentation mapping is applied on a T–F representation, such as log mel spectrogram of an audio clip to obtain T–F segmentation masks of sound events. The T–F segmentation masks can be used for separating the sound events from the background scenes in the T–F domain. Then, a classification mapping is applied on the T–F segmentation masks to estimate the presence probabilities of the sound events. We model the segmentation mapping using a convolutional neural network and the classification mapping using a global weighted rank pooling. In SED, predicted onset and offset times can be obtained from the T–F segmentation masks. As a byproduct, separated waveforms of sound events can be obtained from the T–F segmentation masks. We remixed the DCASE 2018 Task 1 acoustic scene data with the DCASE 2018 Task 2 sound events data. When mixing under 0 dB, the proposed method achieved F1 scores of 0.534, 0.398, and 0.167 in audio tagging, frame-wise SED and event-wise SED, outperforming the fully connected deep neural network baseline of 0.331, 0.237, and 0.120, respectively. In T–F segmentation, we achieved an F1 score of 0.218, where previous methods were not able to do T–F segmentation.

Audio tagging is the task of predicting the presence or absence of sound classes within an audio clip. Previous work in audio tagging focused on relatively small datasets limited to recognizing a small number of sound classes. We investigate audio tagging on AudioSet, which is a dataset consisting of over 2 million audio clips and 527 classes. AudioSet is weakly labelled, in that only the presence or absence of sound classes is known for each clip, whereas the onset and offset times are unknown. To address the weakly labelled audio tagging problem, we propose attention neural networks as a way to attend the most salient parts of an audio clip. We bridge the connection between attention neural networks and multiple instance learning (MIL) methods, and propose decision-level and feature-level attention neural networks for audio tagging. We investigate attention neural networks modeled by different functions, depths, and widths. Experiments on AudioSet show that the feature-level attention neural network achieves a state-of-the-art mean average precision of 0.369, outperforming the best MIL method of 0.317 and Google's deep neural network baseline of 0.314. In addition, we discover that the audio tagging performance on AudioSet-embedding features has a weak correlation with the number of training samples and the quality of labels of each sound class.

In this paper, we present a gated convolutional neural network and a temporal attention-based localization method for audio classification, which won the 1st place in the large-scale weakly supervised sound event detection task of Detection and Classification of Acoustic Scenes and Events (DCASE) 2017 challenge. The audio clips in this task, which are extracted from YouTube videos, are manually labelled with one or more audio tags, but without time stamps of the audio events, hence referred to as weakly labelled data. Two subtasks are defined in this challenge including audio tagging and sound event detection using this weakly labelled data. We propose a convolutional recurrent neural network (CRNN) with learnable gated linear units (GLUs) non-linearity applied on the log Mel spectrogram. In addition, we propose a temporal attention method along the frames to predict the locations of each audio event in a chunk from the weakly labelled data. The performances of our systems were ranked the 1st and the 2nd as a team in these two sub-tasks of DCASE 2017 challenge with F value 55.6% and Equal error 0.73, respectively.

In this paper, we propose a divide-and-conquer approach using two generative adversarial networks (GANs) to explore how a machine can draw colorful pictures (bird) using a small amount of training data. In our work, we simulate the procedure of an artist drawing a picture, where one begins with drawing objects' contours and edges and then paints them different colors. We adopt two GAN models to process basic visual features including shape, texture and color. We use the first GAN model to generate object shape, and then paint the black and white image based on the knowledge learned using the second GAN model. We run our experiments on 600 color images. The experimental results show that the use of our approach can generate good quality synthetic images, comparable to real ones.

This paper investigates the Audio Set classification. Audio Set is a large scale weakly labelled dataset (WLD) of audio clips. In WLD only the presence of a label is known, without knowing the happening time of the labels. We propose an attention model to solve this WLD problem and explain the attention model from a novel probabilistic perspective. Each audio clip in Audio Set consists of a collection of features. We call each feature as an instance and the collection as a bag following the terminology in multiple instance learning. In the attention model, each instance in the bag has a trainable probability measure for each class. The classification of the bag is the expectation of the classification output of the instances in the bag with respect to the learned probability measure. Experiments show that the proposed attention model achieves a mAP of 0.327 on Audio Set, outperforming the Google's baseline of 0.314.

Source separation (SS) aims to separate individual sources from an audio recording. Sound event detection (SED) aims to detect sound events from an audio recording. We propose a joint separation-classification (JSC) model trained only on weakly labelled audio data, that is, only the tags of an audio recording are known but the time of the events are unknown. First, we propose a separation mapping from the time-frequency (T-F) representation of an audio to the T-F segmentation masks of the audio events. Second, a classification mapping is built from each T-F segmentation mask to the presence probability of each audio event. In the source separation stage, sources of audio events and time of sound events can be obtained from the T-F segmentation masks. The proposed method achieves an equal error rate (EER) of 0.14 in SED, outperforming deep neural network baseline of 0.29. Source separation SDR of 8.08 dB is obtained by using global weighted rank pooling (GWRP) as probability mapping, out-performing the global max pooling (GMP) based probability mapping giving SDR at 0.03 dB. Source code of our work is published.

Proximal methods are an important tool in signal processing applications, where many problems can be characterized by the minimization of an expression involving a smooth fitting term and a convex regularization term - for example the classic ℓ1-Lasso. Such problems can be solved using the relevant proximal operator. Here we consider the use of proximal operators for the ℓp-quasinorm where 0 ≤p≤ 1. Rather than seek a closed form solution, we develop an iterative algorithm using a Majorization-Minimization procedure which results in an inexact operator. Experiments on image denoising show that forp≤ 1 the algorithm is effective in the high-noise scenario, outperforming the Lasso despite the inexactness of the proximal step.

Non-negative Matrix Factorization (NMF) is a well established tool for audio analysis. However, it is not well suited for learning on weakly labeled data, i.e. data where the exact timestamp of the sound of interest is not known. In this paper we propose a novel extension to NMF, that allows it to extract meaningful representations from weakly labeled audio data. Recently, a constraint on the activation matrix was proposed to adapt for learning on weak labels. To further improve the method we propose to add an orthogonality regularizer of the dictionary in the cost function of NMF. In that way we obtain appropriate dictionaries for the sounds of interest and background sounds from weakly labeled data. We demonstrate that the proposed Orthogonality-Regularized Masked NMF (ORM-NMF) can be used for Audio Event Detection of rare events and evaluate the method on the development data from Task2 of DCASE2017 Challenge.

There is some uncertainty as to whether objective metrics for predicting the perceived quality of audio source separation are sufficiently accurate. This issue was investigated by employing a revised experimental methodology to collect subjective ratings of sound quality and interference of singing-voice recordings that have been extracted from musical mixtures using state-of-the-art audio source separation. A correlation analysis between the experimental data and the measures of two objective evaluation toolkits, BSS Eval and PEASS, was performed to assess their performance. The artifacts-related perceptual score of the PEASS toolkit had the strongest correlation with the perception of artifacts and distortions caused by singing-voice separation. Both the source-to-interference ratio of BSS Eval and the interference-related perceptual score of PEASS showed comparable correlations with the human ratings of interference.

Radio production involves editing speech-based audio using tools that represent sound using simple waveforms. Semantic speech editing systems allow users to edit audio using an automatically generated transcript, which has the potential to improve the production workflow. To investigate this, we developed a semantic audio editor based on a pilot study. Through a contextual qualitative study of five professional radio producers at the BBC, we examined the existing radio production process and evaluated our semantic editor by using it to create programmes that were later broadcast. We observed that the participants in our study wrote detailed notes about their recordings and used annotation to mark which parts they wanted to use. They collaborated closely with the presenter of their programme to structure the contents and write narrative elements. Participants reported that they often work away from the office to avoid distractions, and print transcripts so they can work away from screens. They also emphasised that listening is an important part of production, to ensure high sound quality. We found that semantic speech editing with automated speech recognition can be used to improve the radio production workflow, but that annotation, collaboration, portability and listening were not well supported by current semantic speech editing systems. In this paper, we make recommendations on how future semantic speech editing systems can better support the requirements of radio production.

Public evaluation campaigns and datasets promote active development in target research areas, allowing direct comparison of algorithms. The second edition of the challenge on detection and classification of acoustic scenes and events (DCASE 2016) has offered such an opportunity for development of the state-of-the-art methods, and succeeded in drawing together a large number of participants from academic and industrial backgrounds. In this paper, we report on the tasks and outcomes of the DCASE 2016 challenge. The challenge comprised four tasks: acoustic scene classification, sound event detection in synthetic audio, sound event detection in real-life audio, and domestic audio tagging. We present each task in detail and analyze the submitted systems in terms of design and performance. We observe the emergence of deep learning as the most popular classification method, replacing the traditional approaches based on Gaussian mixture models and support vector machines. By contrast, feature representations have not changed substantially throughout the years, as mel frequency-based representations predominate in all tasks. The datasets created for and used in DCASE 2016 are publicly available and are a valuable resource for further research.

Musical noise is a recurrent issue that appears in spectral techniques for denoising or blind source separation. Due to localised errors of estimation, isolated peaks may appear in the processed spectrograms, resulting in annoying tonal sounds after synthesis known as “musical noise”. In this paper, we propose a method to assess the amount of musical noise in an audio signal, by characterising the impact of these artificial isolated peaks on the processed sound. It turns out that because of the constraints between STFT coefficients, the isolated peaks are described as time-frequency “spots” in the spectrogram of the processed audio signal. The quantification of these “spots”, achieved through the adaptation of a method for localisation of significant STFT regions, allows for an evaluation of the amount of musical noise. We believe that this will pave the way to an objective measure and a better understanding of this phenomenon.

Automatic and fast tagging of natural sounds in audio collections is a very challenging task due to wide acoustic variations, the large number of possible tags, the incomplete and ambiguous tags provided by different labellers. To handle these problems, we use a co-regularization approach to learn a pair of classifiers on sound and text. The first classifier maps low-level audio features to a true tag list. The second classifier maps actively corrupted tags to the true tags, reducing incorrect mappings caused by low-level acoustic variations in the first classifier, and to augment the tags with additional relevant tags. Training the classifiers is implemented using marginal co-regularization, pair of which draws the two classifiers into agreement by a joint optimization. We evaluate this approach on two sound datasets, Freefield1010 and Task4 of DCASE2016. The results obtained show that marginal co-regularization outperforms the baseline GMM in both efficiency and effectiveness.

Audio tagging aims to assign one or several tags to an audio clip. Most of the datasets are weakly labelled, which means only the tags of the clip are known, without knowing the occurrence time of the tags. The labeling of an audio clip is often based on the audio events in the clip and no event level label is provided to the user. Previous works have used the bag of frames model assume the tags occur all the time, which is not the case in practice. We propose a joint detection-classification (JDC) model to detect and classify the audio clip simultaneously. The JDC model has the ability to attend to informative and ignore uninformative sounds. Then only informative regions are used for classification. Experimental results on the “CHiME Home” dataset show that the JDC model reduces the equal error rate (EER) from 19.0% to 16.9%. More interestingly, the audio event detector is trained successfully without needing the event level label.

We address the problem of sparse signal reconstruction from a few noisy samples. Recently, a Covariance-Assisted Matching Pursuit (CAMP) algorithm has been proposed, improving the sparse coefficient update step of the classic Orthogonal Matching Pursuit (OMP) algorithm. CAMP allows the a-priori mean and covariance of the non-zero coefficients to be considered in the coefficient update step. In this paper, we analyze CAMP, which leads to a new interpretation of the update step as a maximum-a-posteriori (MAP) estimation of the non-zero coefficients at each step. We then propose to leverage this idea, by finding a MAP estimate of the sparse reconstruction problem, in a greedy OMP-like way. Our approach allows the statistical dependencies between sparse coefficients to be modelled, while keeping the practicality of OMP. Experiments show improved performance when reconstructing the signal from a few noisy samples.

Environmental audio tagging is a newly proposed task to predict the presence or absence of a specific audio event in a chunk. Deep neural network (DNN) based methods have been successfully adopted for predicting the audio tags in the domestic audio scene. In this paper, we propose to use a convolutional neural network (CNN) to extract robust features from mel-filter banks (MFBs), spectrograms or even raw waveforms for audio tagging. Gated recurrent unit (GRU) based recurrent neural networks (RNNs) are then cascaded to model the long-term temporal structure of the audio signal. To complement the input information, an auxiliary CNN is designed to learn on the spatial features of stereo recordings. We evaluate our proposed methods on Task 4 (audio tagging) of the Detection and Classification of Acoustic Scenes and Events 2016 (DCASE 2016) challenge. Compared with our recent DNN-based method, the proposed structure can reduce the equal error rate (EER) from 0.13 to 0.11 on the development set. The spatial features can further reduce the EER to 0.10. The performance of the end-to-end learning on raw waveforms is also comparable. Finally, on the evaluation set, we get the state-of-the-art performance with 0.12 EER while the performance of the best existing system is 0.15 EER.

The first generation of three-dimensional Electromagnetic Articulography \ndevices (Carstens AG500) suffered from occasional \ncritical tracking failures. Although now superseded by \nnew devices, the AG500 is still in use in many speech labs \nand many valuable data sets exist. In this study we investigate \nwhether deep neural networks (DNNs) can learn the mapping \nfunction from raw voltage amplitudes to sensor positions based \non a comprehensive movement data set. This is compared to \narriving sample by sample at individual position values via direct \noptimisation as used in previous methods. We found that \nwith appropriate hyperparameter settings a DNN was able to \napproximate the mapping function with good accuracy, leading \nto a smaller error than the previous methods, but that the \nDNN-based approach was not able to solve the tracking problem \ncompletely.

Audio tagging aims to perform multi-label classification on audio \nchunks and it is a newly proposed task in the Detection and \nClassification of Acoustic Scenes and Events 2016 (DCASE \n2016) challenge. This task encourages research efforts to better \nanalyze and understand the content of the huge amounts of \naudio data on the web. The difficulty in audio tagging is that \nit only has a chunk-level label without a frame-level label. This \npaper presents a weakly supervised method to not only predict \nthe tags but also indicate the temporal locations of the occurred \nacoustic events. The attention scheme is found to be effective \nin identifying the important frames while ignoring the unrelated \nframes. The proposed framework is a deep convolutional recurrent \nmodel with two auxiliary modules: an attention module \nand a localization module. The proposed algorithm was evaluated \non the Task 4 of DCASE 2016 challenge. State-of-the-art \nperformance was achieved on the evaluation set with equal error \nrate (EER) reduced from 0.13 to 0.11, compared with the \nconvolutional recurrent baseline system.

Binaural features of interaural level difference and interaural phase difference have proved to be very effective in training deep neural networks (DNNs), to generate time-frequency masks for target speech extraction in speech-speech mixtures. However, effectiveness of binaural features is reduced in more common speech-noise scenarios, since the noise may over-shadow the speech in adverse conditions. In addition, the reverberation also decreases the sparsity of binaural features and therefore adds difficulties to the separation task. To address the above limitations, we highlight the spectral difference between speech and noise spectra and incorporate the log-power spectra features to extend the DNN input. Tested on two different reverberant rooms at different signal to noise ratios (SNR), our proposed method shows advantages over the baseline method using only binaural features in terms of signal to distortion ratio (SDR) and Short-Time Perceptual Intelligibility (STOI).

In this paper, a system for polyphonic sound event detection and tracking is proposed, based on spectrogram factorization techniques and state space models. The system extends probabilistic latent component analysis (PLCA) and is modeled around a four-dimensional spectral template dictionary of frequency, sound event class, exemplar index, and sound state. In order to jointly track multiple overlapping sound events over time, the integration of linear dynamical systems (LDS) within the PLCA inference is proposed. The system assumes that the PLCA sound event activation is the (noisy) observation in an LDS, with the latent states corresponding to the true event activations. LDS training is achieved using fully observed data, making use of ground truth-informed event activations produced by the PLCA-based model. Several LDS variants are evaluated, using polyphonic datasets of office sounds generated from an acoustic scene simulator, as well as real and synthesized monophonic datasets for comparative purposes. Results show that the integration of LDS tracking within PLCA leads to an improvement of +8.5-10.5% in terms of frame-based F-measure as compared to the use of the PLCA model alone. In addition, the proposed system outperforms several state-of-the-art methods for the task of polyphonic sound event detection.

Most single channel audio source separation approaches produce separated sources accompanied by interference from other sources and other distortions. To tackle this problem, we propose to separate the sources in two stages. In the first stage, the sources are separated from the mixed signal. In the second stage, the interference between the separated sources and the distortions are reduced using deep neural networks (DNNs). We propose two methods that use DNNs to improve the quality of the separated sources in the second stage. In the first method, each separated source is improved individually using its own trained DNN, while in the second method all the separated sources are improved together using a single DNN. To further improve the quality of the separated sources, the DNNs in the second stage are trained discriminatively to further decrease the interference and the distortions of the separated sources. Our experimental results show that using two stages of separation improves the quality of the separated signals by decreasing the interference between the separated sources and distortions compared to separating the sources using a single stage of separation.

Environmental audio tagging aims to predict only the presence or absence of certain acoustic events in the interested acoustic scene. In this paper, we make contributions to audio tagging in two parts, respectively, acoustic modeling and feature learning. We propose to use a shrinking deep neural network (DNN) framework incorporating unsupervised feature learning to handle the multilabel classification task. For the acoustic modeling, a large set of contextual frames of the chunk are fed into the DNN to perform a multilabel classification for the expected tags, considering that only chunk (or utterance) level rather than frame-level labels are available. Dropout and background noise aware training are also adopted to improve the generalization capability of the DNNs. For the unsupervised feature learning, we propose to use a symmetric or asymmetric deep denoising auto-encoder (syDAE or asyDAE) to generate new data-driven features from the logarithmic Mel-filter banks features. The new features, which are smoothed against background noise and more compact with contextual information, can further improve the performance of the DNN baseline. Compared with the standard Gaussian mixture model baseline of the DCASE 2016 audio tagging challenge, our proposed method obtains a significant equal error rate (EER) reduction from 0.21 to 0.13 on the development set. The proposed asyDAE system can get a relative 6.7% EER reduction compared with the strong DNN baseline on the development set. Finally, the results also show that our approach obtains the state-of-the-art performance with 0.15 EER on the evaluation set of the DCASE 2016 audio tagging task while EER of the first prize of this challenge is 0.17.

In this paper, a system for overlapping acoustic event detection is proposed, which models the temporal evolution of sound events. The system is based on probabilistic latent component analysis, supporting the use of a sound event dictionary where each exemplar consists of a succession of spectral templates. The temporal succession of the templates is controlled through event class-wise Hidden Markov Models (HMMs). As input time/frequency representation, the Equivalent Rectangular Bandwidth (ERB) spectrogram is used. Experiments are carried out on polyphonic datasets of office sounds generated using an acoustic scene simulator, as well as real and synthesized monophonic datasets for comparative purposes. Results show that the proposed system outperforms several state-of-the-art methods for overlapping acoustic event detection on the same task, using both frame-based and event-based metrics, and is robust to varying event density and noise levels.

Deep neural networks (DNNs) are usually used for single channel source separation to predict either soft or binary time frequency masks. The masks are used to separate the sources from the mixed signal. Binary masks produce separated sources with more distortion and less interference than soft masks. In this paper, we propose to use another DNN to combine the estimates of binary and soft masks to achieve the advantages and avoid the disadvantages of using each mask individually. We aim to achieve separated sources with low distortion and low interference between each other. Our experimental results show that combining the estimates of binary and soft masks using DNN achieves lower distortion than using each estimate individually and achieves as low interference as the binary mask.

Automatic music transcription (AMT) can be performed by deriving a pitch-time representation through decomposition of a spectrogram with a dictionary of pitch-labelled atoms. Typically, non-negative matrix factorisation (NMF) methods are used to decompose magnitude spectrograms. One atom is often used to represent each note. However, the spectrum of a note may change over time. Previous research considered this variability using different atoms to model specific parts of a note, or large dictionaries comprised of datapoints from the spectrograms of full notes. In this paper, the use of subspace modelling of note spectra is explored, with group sparsity employed as a means of coupling activations of related atoms into a pitched subspace. Stepwise and gradient-based methods for non-negative group sparse decompositions are proposed. Finally, a group sparse NMF approach is used to tune a generic harmonic subspace dictionary, leading to improved NMF-based AMT results.

In the context of the Internet of Things, sound sensing applications are required to run on embedded platforms where notions of product pricing and form factor impose hard constraints on the available computing power. Whereas Automatic Environmental Sound Recognition (AESR) algorithms are most often developed with limited consideration for computational cost, this paper seeks which AESR algorithm can make the most of a limited amount of computing power by comparing the sound classification performance as a function of its computational cost. Results suggest that Deep Neural Networks yield the best ratio of sound classification accuracy across a range of computational costs, while Gaussian Mixture Models offer a reasonable accuracy at a consistently small cost, and Support Vector Machines stand between both in terms of compromise between accuracy and computational cost.

Given a musical audio recording, the goal of music transcription is to determine a score-like representation of the piece underlying the recording. Most current transcription methods employ variants of non-negative matrix factorization (NMF), which often fails to robustly model instruments producing non-stationary sounds. Using entire time-frequency patterns to represent sounds, non-negative matrix deconvolution (NMD) can capture certain types of non-stationary behavior but is only applicable if all sounds have the same length. In this paper, we present a novel method that combines the non-stationarity modeling capabilities available with NMD with the variable note lengths possible with NMF. Identifying frames in NMD patterns with states in a dynamical system, our method iteratively generates sound-object candidates separately for each pitch, which are then combined in a global optimization. We demonstrate the transcription capabilities of our method using piano pieces assuming the availability of single note recordings as training data.

Non-negative Matrix Factorisation (NMF) is a commonly used tool in many musical signal processing tasks, including Automatic Music Transcription (AMT). However unsupervised NMF is seen to be problematic in this context, and harmonically constrained variants of NMF have been proposed. While useful, the harmonic constraints may be constrictive in mixed signals. We have previously observed that recovery of overlapping signal elements using NMF is improved through introduction of a sparse coding step, and propose here the incorporation of a sparse coding step using the Hellinger distance into a NMF algorithm. Improved AMT results for unsupervised NMF are reported.

For intelligent systems to make best use of the audio modality, it is important that they can recognize not just speech and music, which have been researched as specific tasks, but also general sounds in everyday environments. To stimulate research in this field we conducted a public research challenge: the IEEE Audio and Acoustic Signal Processing Technical Committee challenge on Detection and Classification of Acoustic Scenes and Events (DCASE). In this paper, we report on the state of the art in automatically classifying audio scenes, and automatically detecting and classifying audio events. We survey prior work as well as the state of the art represented by the submissions to the challenge from various research groups. We also provide detail on the organization of the challenge, so that our experience as challenge hosts may be useful to those organizing challenges in similar domains. We created new audio datasets and baseline systems for the challenge; these, as well as some submitted systems, are publicly available under open licenses, to serve as benchmarks for further research in general-purpose machine listening.

Techniques based on non-negative matrix factorization (NMF) have been successfully used to decompose a spectrogram of a music recording into a dictionary of templates and activations. While advanced NMF variants often yield robust signal models, there are usually some inaccuracies in the factorization since the underlying methods are not prepared for phase cancellations that occur when sounds with similar frequency are mixed. In this paper, we present a novel method that takes phase cancellations into account to refine dictionaries learned by NMF-based methods. Our approach exploits the fact that advanced NMF methods are often robust enough to provide information about how sound sources interact in a spectrogram, where they overlap, and thus where phase cancellations could occur. Using this information, the distances used in NMF are weighted entry-wise to attenuate the influence of regions with phase cancellations. Experiments on full-length, polyphonic piano recordings indicate that our method can be successfully used to refine NMF-based dictionaries.

A method is proposed for instrument recognition in polyphonic music which combines two independent detector systems. A polyphonic musical instrument recognition system using a missing feature approach and an automatic music transcription system based on shift invariant probabilistic latent component analysis that includes instrument assignment. We propose a method to integrate the two systems by fusing the instrument contributions estimated by the first system onto the transcription system in the form of Dirichlet priors. Both systems, as well as the integrated system are evaluated using a dataset of continuous polyphonic music recordings. Detailed results that highlight a clear improvement in the performance of the integrated system are reported for different training conditions.

Non-negative Matrix Factorisation (NMF) is a popular tool in musical signal processing. However, problems using this methodology in the context of Automatic Music Transcription (AMT) have been noted resulting in the proposal of supervised and constrained variants of NMF for this purpose. Group sparsity has previously been seen to be effective for AMT when used with stepwise methods. In this paper group sparsity is introduced to supervised NMF decompositions and a dictionary tuning approach to AMT is proposed based upon group sparse NMF using the β-divergence. Experimental results are given showing improved AMT results over the state-of-the-art NMF-based AMT system.

1628

Several probabilistic models involving latent components have been proposed for modeling time-frequency (TF) representations of audio signals such as spectrograms, notably in the nonnegative matrix factorization (NMF) literature. Among them, the recent high-resolution NMF (HR-NMF) model is able to take both phases and local correlations in each frequency band into account, and its potential has been illustrated in applications such as source separation and audio inpainting. In this paper, HR-NMF is extended to multichannel signals and to convolutive mixtures. The new model can represent a variety of stationary and non-stationary signals, including autoregressive moving average (ARMA) processes and mixtures of damped sinusoids. A fast variational expectation-maximization (EM) algorithm is proposed to estimate the enhanced model. This algorithm is applied to piano signals, and proves capable of accurately modeling reverberation, restoring missing observations, and separating pure tones with close frequencies.

In this paper we present a new method for musical audio source separation, using the information from the musical score to supervise the decomposition process. An original framework using nonnegative matrix factorization (NMF) is presented, where the components are initially learnt on synthetic signals with temporal and harmonic constraints. A new dataset of multitrack recordings with manually aligned MIDI scores is created (TRIOS), and we compare our separation results with other methods from the literature using the BSS EVAL and PEASS evaluation toolboxes. The results show a general improvement of the BSS EVAL metrics for the various instrumental configurations used.

A method based on local spectral features and missing feature techniques is proposed for the recognition of harmonic sounds in mixture signals. A mask estimation algorithm is proposed for identifying spectral regions that contain reliable information for each sound source and then bounded marginalization is employed to treat the feature vector elements that are determined as unreliable. The proposed method is tested on musical instrument sounds due to the extensive availability of data but it can be applied on other sounds (i.e. animal sounds, environmental sounds), whenever these are harmonic. In simulations the proposed method clearly outperformed a baseline method for mixture signals.

In this work, we study the links between the recovery properties of sparse signals for Orthogonal Matching Pursuit (OMP) and the whole General MP class over nested supports. We show that the optimality of those algorithms is not locally nested: there is a dictionary and supports I and J with J included in I such that OMP will recover all signals of support I, but not all signals of support J. We also show that the optimality of OMP is globally nested: if OMP can recover all s-sparse signals, then it can recover all s'-sparse signals with s' smaller than s. We also provide a tighter version of Donoho and Elad's spark theorem, which allows us to complete Tropp's proof that sparse representation algorithms can only be optimal for all s-sparse signals if s is strictly lower than half the spark of the dictionary.

Automatic Music Transcription (AMT) seeks to understand a musical piece in terms of note activities. Matrix decomposition methods are often used for AMT, seeking to decompose a spectrogram over a dictionary matrix of note-specific template vectors. The performance of these methods can suffer due to the large harmonic overlap found in tonal musical spectra. We propose a row weighting scheme that transforms each spectrogram frame and the dictionary, with the weighting determined by the effective correlations in the decomposition. Experiments show improved AMT performance.

Segregating an audio mixture containing multiple simultaneous bird sounds is a challenging task. However, birdsong often contains rapid pitch modulations, and these modulations carry information which may be of use in automatic recognition. In this paper we demonstrate that an improved spectrogram representation, based on the distribution derivative method, leads to improved performance of a segregation algorithm which uses a Markov renewal process model to track vocalisation patterns consisting of singing and silences.

It is shown that Theorem 10 (Non-Nestedness of ERC) in [Plumbley, IEEE Trans. Inf. Theory, vol. 53, pp. 3188-3195, Sep. 2007] neglects the derivations of the exact recovery conditions (ERCs) of constrained l1-minimization (BP) and orthogonal matching pursuit. This means that it does not reflect the recovery properties of these algorithms. Furthermore, an ERC of BP more general than that in [Tropp, IEEE Trans. Inf. Theory, vol. 50, pp. 2231-2242, Oct. 2004] is shown.

Although researchers are increasingly aware of the need to publish and maintain software code alongside their results, practical barriers prevent this from happening in many cases. We examine these barriers, propose an incremental approach to overcoming some of them, and describe the Sound Software project, an effort to support software development practice in the UK audio and music research community. Finally we make some recommendations for research groups seeking to improve their own researchers' software practice.

Analysis based reconstruction has recently been introduced as an alternative to the well-known synthesis sparsity model used in a variety of signal processing areas. In this paper we convert the analysis exact-sparse reconstruction problem to an equivalent synthesis recovery problem with a set of additional constraints. We are therefore able to use existing synthesis-based algorithms for analysis-based exact-sparse recovery. We call this the Analysis-By-Synthesis (ABS) approach. We evaluate our proposed approach by comparing it against the recent Greedy Analysis Pursuit (GAP) analysis-based recovery algorithm. The results show that our approach is a viable option for analysis-based reconstruction, while at the same time allowing many algorithms that have been developed for synthesis reconstruction to be directly applied for analysis reconstruction as well.

This paper describes a novel music similarity calculation method that is based on the instrumentation of music pieces. The approach taken here is based on the idea that sparse representations of musical audio signals are a rich source of information regarding the elements that constitute the observed spectra. We propose a method to extract feature vectors based on sparse representations and use these to calculate a similarity measure between songs. To train a dictionary for sparse representations from a large amount of training data, a novel dictionary-initialization method based on agglomerative clustering is proposed. An objective evaluation shows that the new features improve the performance of similarity calculation compared to the standard mel-frequency cepstral coefficients features.

This work considers the problem of learning an incoherent dictionary that is both adapted to a set of training data and incoherent so that existing sparse approximation algorithms can recover the sparsest representation. A new decorrelation method is presented that computes a fixed coherence dictionary close to a given dictionary. That step iterates pairwise decorrelations of atoms in the dictionary. Dictionary learning is then performed by adding this decorrelation method as an intermediate step in the K-SVD learning algorithm. The proposed algorithm INK-SVD is tested on musical data and compared to another existing decorrelation method. INK-SVD can compute a dictionary that approximates the training data as well as K-SVD while decreasing the coherence from 0.6 to 0.2.

Sparse representations have previously been applied to the automatic music transcription (AMT) problem. Structured sparsity, such as group and molecular sparsity allows the introduction of prior knowledge to sparse representations. Molecular sparsity has previously been proposed for AMT, however the use of greedy group sparsity has not previously been proposed for this problem. We propose a greedy sparse pursuit based on nearest subspace classification for groups with coherent blocks, based in a non-negative framework, and apply this to AMT. Further to this, we propose an enhanced molecular variant of this group sparse algorithm and demonstrate the effectiveness of this approach.

We propose the audio inpainting framework that recovers portions of audio data distorted due to impairments such as impulsive noise, clipping, and packet loss. In this framework, the distorted data are treated as missing and their location is assumed to be known. The signal is decomposed into overlapping time-domain frames and the restoration problem is then formulated as an inverse problem per audio frame. Sparse representation modeling is employed per frame, and each inverse problem is solved using the Orthogonal Matching Pursuit algorithm together with a discrete cosine or a Gabor dictionary. The Signal-to-Noise Ratio performance of this algorithm is shown to be comparable or better than state-of-the-art methods when blocks of samples of variable durations are missing. We also demonstrate that the size of the block of missing samples, rather than the overall number of missing samples, is a crucial parameter for high quality signal restoration. We further introduce a constrained Matching Pursuit approach for the special case of audio declipping that exploits the sign pattern of clipped audio samples and their maximal absolute value, as well as allowing the user to specify the maximum amplitude of the signal. This approach is shown to outperform state-of-the-art and commercially available methods for audio declipping in terms of Signal-to-Noise Ratio.

A new probabilistic framework for beat tracking of musical audio is presented. The method estimates the time between consecutive beat events and exploits both beat and non-beat information by explicitly modeling non-beat states. In addition to the beat times, a measure of the expected accuracy of the estimated beats is provided. The quality of the observations used for beat tracking is measured and the reliability of the beats is automatically calculated. Ak-nearest neighbor regression algorithm is proposed to predict the accuracy of the beat estimates. The performance of the beat tracking system is statistically evaluated using a database of 222 musical signals of various genres. We show that modeling non-beat states leads to a significant increase in performance. In addition, a large experiment where the parameters of the model are automatically learned has been completed. Results show that simple approximations for the parameters of the model can be used. Furthermore, the performance of the system is compared with existing algorithms. Finally, a new perspective for beat tracking evaluation is presented. We show how reliability information can be successfully used to increase the mean performance of the proposed algorithm and discuss how far automatic beat tracking is from human tapping.

This paper introduces a technique for predicting harmonic sequences in a musical performance for which no score is available, using real-time audio signals. Recent short-term information is aligned with longer term information, contextualizing the present within the past, allowing predictions about the future of the performance to be made. Using a mid-level representation in the form of beat-synchronous harmonic sequences, we reduce the size of the information needed to represent the performance. This allows the implementation of real-time performance following in live performance situations. We conduct an objective evaluation on a database of rock, pop, and folk music. Our results show that we are able to predict a large majority of repeated harmonic content with no prior knowledge in the form of a score.

We present a novel sparse representation based approach for the restoration of clipped audio signals. In the proposed approach, the clipped signal is decomposed into overlapping frames and the declipping problem is formulated as an inverse problem, per audio frame. This problem is further solved by a constrained matching pursuit algorithm, that exploits the sign pattern of the clipped samples and their maximal absolute value. Performance evaluation with a collection of music and speech signals demonstrate superior results compared to existing algorithms, over a wide range of clipping levels.

Assuming that a set of source signals is sparsely representable in a given dictionary, we show how their sparse recovery fails whenever we can only measure a convolved observation of them. Starting from this motivation, we develop a block coordinate descent method which aims to learn a convolved dictionary and provide a sparse representation of the observed signals with small residual norm. We compare the proposed approach to the K-SVD dictionary learning algorithm and show through numerical experiment on synthetic signals that, provided some conditions on the problem data, our technique converges in a fixed number of iterations to a sparse representation with smaller residual norm.

In this paper, we consider the problem of separating a set of independent components when only one movable sensor is available to record the mixtures. We propose to exploit the quasi-periodicity of the heart signals to transform the signal from this one moving sensor, into a set of measurements, as if from a virtual array of sensors. We then use ICA to perform source separation. We show that this technique can be applied to heart sounds and to electrocardiograms.

We present a new evaluation method for measuring the performance of musical audio beat tracking systems. Central to our method is a novel visualization, the beat error histogram, which illustrates the metrical relationship between two qausi-periodic sequences of time instants: the output of beat tracking system and a set of ground truth annotations. To quantify beat tracking performance we derive an information theoretic statistic from the histogram. Results indicate that our method is able to measure performance with greater precision than existing evaluation methods and implicitly cater for metrical ambiguity in tapping sequences.

In this paper we explore the relationship between the temporal and rhythmic structure of musical audio signals. Using automatically extracted rhythmic structure we present a rhythmically-aware method to combine note onset detection techniques. Our method uses top-down knowledge of repetitions of musical events to improve detection performance by modelling the temporal distribution of onset locations. Results on a publicly available database demonstrate that using musical knowledge in this way can lead to significant improvements by reducing the number of missed and spurious detections.

Polytope Faces Pursuit is a greedy algorithm that performs Basis Pursuit with similar order complexity to Orthogonal Matching Pursuit. The algorithm adds one basis vector at a time and adopts a path-following approach based on the geometry of the polar polytope associated with the dual Linear Program. Its initial implementation uses the method of Cholesky factorization to update the solution vector at each step, which can be computationally expensive for solving large scale problems as it requires the succesive storage of large matrices. In this paper, we present a different approach using directional updates to estimate the solution vector at each time. The proposed method uses the gradient descent method, reducing the memory requirements and computational complexity. We demonstrate the application of this Gradient Polytope Faces Pursuit algorithm to a source separation problem.

We propose an ℓ1criterion for dictionary learning for sparse signal representation. Instead of directly searching for the dictionary vectors, our dictionary learning approach identifies vectors that are orthogonal to the subspaces in which the training data concentrate. We study conditions on the coefficients of training data that guarantee that ideal normal vectors deduced from the dictionary are local optima of the criterion. We illustrate the behavior of the criterion on a 2D example, showing that the local minima correspond to ideal normal vectors when the number of training data is sufficient. We conclude by describing an algorithm that can be used to optimize the criterion in higher dimension.

We present a technique for following a live performance in the situation where a score is not available. Making use of a local alignment between recent and longer term musical information, we place the present in the context of the past, allowing the prediction of future performance information. By representing music as sequences of beat-synchronous features we reduce the size of the information needed to represent the performance and allow performance following in real-time to occur.

Sparse representations have proved a powerful tool in the analysis and processing of audio signals and already lie at the heart of popular coding standards such as MP3 and Dolby AAC. In this paper we give an overview of a number of current and emerging applications of sparse representations in areas from audio coding, audio enhancement and music transcription to blind source separation solutions that can solve the “cocktail party problem.” In each case we will show how the prior assumption that the audio signals are approximately sparse in some time-frequency representation allows us to address the associated signal processing task.

We have implemented several fast and flexible adaptive lapped orthogonal transform (LOT) schemes for underdetermined audio source separation. This is generally addressed by time-frequency masking, requiring the sources to be disjoint in the time-frequency domain. We have already shown that disjointness can be increased via adaptive dyadic LOTs. By taking inspiration from the windowing schemes used in many audio coding frameworks, we improve on earlier results in two ways. Firstly, we consider non-dyadic LOTs which match the time-varying signal structures better. Secondly, we allow for a greater range of overlapping window profiles to decrease window boundary artifacts. This new scheme is benchmarked through oracle evaluations, and is shown to decrease computation time by over an order of magnitude compared to using very general schemes, whilst maintaining high separation performance and flexible signal adaptivity. As the results demonstrate, this work may find practical applications in high fidelity audio source separation.

This paper describes a method for solving the permutation problem in the frequency-domain independent component analysis (FD-ICA) approach to blind source separation (BSS). FD-ICA is a well-known method for BSS of convolutive mixtures. However, FD-ICA has a source permutation problem, where estimated source components can become swapped at different frequencies. Many researchers have suggested methods to solve the source permutation problem including using correlation between adjacent frequencies and direction of arrival (DOA). We propose a modification to the DOA method, based on phase linearity of the FD-ICA de-mixing matrix, that can extend the range of frequencies over which the permutation problem can be resolved. Experiments indicate that our method can provide a better performance than the inter-frequency correlation method and the DOA method in real environments.

Measures such as entropy and mutual information can be used to characterise random processes. In this paper, we propose the use of several time-varying information measures, computed in the context of a probabilistic model that evolves as a sample of the process unfolds, as a way to characterise temporal structure in music. One such measure is a novel predictive information rate which we conjecture may provide a conceptually simple explanation for the ‘inverted-U’ relationship often found between simple measures of randomness (e.g. entropy rate) and judgements of aesthetic value [Berlyne, D.E. (1971), Aesthetics and Psychobiology, New York: Appleton Century Crofts.]. We explore these ideas in the context of Markov chains using both artificially generated sequences and two pieces of minimalist music by Philip Glass, showing that even such a manifestly simplistic model (the Markov chain), when interpreted according to information dynamic principles, produces a structural analysis which largely agrees with that of an expert human listener. Thus, we propose that our approach could form the basis of a theoretically coherent yet computationally plausible model of human perception of formal structure, potentially including seemingly abstract qualities like interestingness and aesthetic goodness.

We describe a nonparametric estimator for the differential entropy of a multidimensional distribution, given a limited set of data points, by a recursive rectilinear partitioning. The estimator uses an adaptive partitioning method and runs in Theta(NlogN) time, with low memory requirements. In experiments using known distributions, the estimator is several orders of magnitude faster than other estimators, with only modest increase in bias and variance.

We address the problem of instantaneous, underdetermined audio source separation by time-frequency masking. Using oracle estimators, we determine experimental upper performance bounds, by assuming that we have reference sources available, and that we know, or have estimated, the mixing structure. Oracle estimation of four musical sources from two-channel mixtures demonstrates a potential for SDR improvements of up to 12.7 dB, compared to semi-blind methods. We also show that using adaptive cosine packet transforms, rather than fixed-basis STFTs, can improve performance by up to 2.2 dB. Finally, by allowing more than one non-zero source coefficient per time-frequency index, improvements of up to 7.7 dB could be possible.

Despite continued attention toward the problem of automatic beat detection in musical audio, the issue of how to evaluate beat tracking systems remains pertinent and controversial. As yet no consistent evaluation metric has been adopted by the research community. To this aim, we propose a new method for beat tracking evaluation by measuring beat accuracy in terms of the entropy of a beat error histogram. We demonstrate the ability of our approach to address several shortcomings of existing methods.

We investigate the use of the Riemannian optimization method over the flag manifold in subspace ICA problems such as independent subspace analysis (ISA) and complex ICA. In the ISA experiment, we use the Riemannian approach over the flag manifold together with an MCMC method to overcome the problem of local minima of the ISA cost function. Experiments demonstrate the effectiveness of both Riemannian methods - simple geodesic gradient descent and hybrid geodesic gradient descent, compared with the ordinary gradient method.

In the last few years, there has been a great interest in the use of geometrical methods for independent component analysis (ICA), both to gain insight into the optimization process and to develop more efficient optimization algorithms. Much of this work involves concepts from differential geometry, such as Lie groups, Stiefel manifolds, or tangent planes that may be unfamiliar to signal processing researchers. The purpose of this tutorial paper is to introduce some of these geometry concepts to signal processing and ICA researchers, without assuming any existing background in differential geometry. The emphasis of the paper is on making the important concepts in this field accessible, rather than mathematical rigour.

We present a simple and efficient method for beat tracking of musical audio. With the aim of replicating the human ability of tapping in time to music, we formulate our approach using a two state model. The first state performs tempo induction and tracks tempo changes, while the second maintains contextual continuity within a single tempo hypothesis. Beat times are recovered by passing the output of an onset detection function through adaptively weighted comb filterbank matrices to separately identify the beat period and alignment. We evaluate our beat tracker both in terms of the accuracy of estimated beat locations and computational complexity. In a direct comparison with existing algorithms, we demonstrate equivalent performance at significantly reduced computational cost

This paper deals with the decomposition of music signals into pitched sound objects made of harmonic sinusoidal partials for very low bit-rate coding purposes. After a brief review of existing methods, we recast this problem in the Bayesian framework. We propose a family of probabilistic signal models combining learned object priors and various perceptually motivated distortion measures. We design efficient algorithms to infer object parameters and build a coder based on the interpolation of frequency and amplitude parameters. Listening tests suggest that the loudness-based distortion measure outperforms other distortion measures and that our coder results in a better sound quality than baseline transform and parametric coders at 8 and 2 kbit/s. This work constitutes a new step towards a fully object-based coding system, which would represent audio signals as collections of meaningful note-like sound objects

Suppose we have a signal${\mmb y}$which we wish to represent using a linear combination of a number of basis atoms$\break {\mmb a}\!\!\!$$_i,{\mmb y}$$=\!\!\!\sum_i x_i {\mmb a}_i ={\mmb A}{\mmb x}$. The problem of finding the minimum$\ell_0$norm representation for${\mmb y}$is a hard problem. The Basis Pursuit (BP) approach proposes to find the minimum$\ell_1$norm representation instead, which corresponds to a linear program (LP) that can be solved using modern LP techniques, and several recent authors have given conditions for the BP (minimum$\ell_1$norm) and sparse (minimum$\ell_0$norm) representations to be identical. In this paper, we explore thissparse representation problemusing the geometry of convex polytopes, as recently introduced into the field by Donoho. By considering the dual LP we find that the so-called polar polytope$P^*$of the centrally symmetric polytope$P$whose vertices are the atom pairs${\pm}{\mmb a}$$_i$is particularly helpful in providing us with geometrical insight into optimality conditions given by Fuchs and Tropp for non-unit-norm atom sets. In exploring this geome

We investigate a data-driven approach to the analysis and transcription of polyphonic music, using a probabilistic model which is able to find sparse linear decompositions of a sequence of short-term Fourier spectra. The resulting system represents each input spectrum as a weighted sum of a small number of "atomic" spectra chosen from a larger dictionary; this dictionary is, in turn, learned from the data in such a way as to represent the given training set in an (information theoretically) efficient way. When exposed to examples of polyphonic music, most of the dictionary elements take on the spectral characteristics of individual notes in the music, so that the sparse decomposition can be used to identify the notes in a polyphonic mixture. Our approach differs from other methods of polyphonic analysis based on spectral decomposition by combining all of the following: (a) a formulation in terms of an explicitly given probabilistic model, in which the process estimating which notes are present corresponds naturally with the inference of latent variables in the model; (b) a particularly simple generative model, motivated by very general considerations about efficient coding, that makes very few assumptions about the musical origins of the signals being processed; and (c) the ability to learn a dictionary of atomic spectra (most of which converge to harmonic spectral profiles associated with specific notes) from polyphonic examples alone-no separate training on monophonic examples is required.

In this paper we apply a two state switching model to the problem of audio based beat tracking. Our analysis is based around the generation and application of adaptively weighted comb filterbank structures to extract beat timing information from the midlevel representation of an input audio signal known as the onset detection function. We evaluate our system using a previously published dataset (S. Hainsworth, Ph.D. thesis, Cambridge Univ., 2004), and in performing a direct comparison with the current state of the art, present comparable results.

The instantaneous noise-free linear mixing model in independent component analysis is largely a solved problem under the usual assumption of independent nongaussian sources and full column rank mixing matrix. However, with some prior information on the sources, like positivity, new analysis and perhaps simplified solution methods may yet become possible. In this letter, we consider the task of independent component analysis when the independent sources are known to be nonnegative and well grounded, which means that they have a nonzero pdf in the region of zero. It can be shown that in this case, the solution method is basically very simple: an orthogonal rotation of the whitened observation vector into nonnegative outputs will give a positive permutation of the original sources. We propose a cost function whose minimum coincides with nonnegativity and derive the gradient algorithm under the whitening constraint, under which the separating matrix is orthogonal. We further prove that in the Stiefel manifold of orthogonal matrices, the cost function is a Lyapunov function for the matrix gradient flow, implying global convergence. Thus, this algorithm is guaranteed to find the nonnegative well-grounded independent sources. The analysis is complemented by a numerical simulation, which illustrates the algorithm.

We consider the task of independent component analysis when the independent sources are known to be nonnegative and well-grounded, so that they have a nonzero probability density function (pdf) in the region of zero. We propose the use of a "nonnegative principal component analysis (nonnegative PCA)" algorithm, which is a special case of the nonlinear PCA algorithm, but with a rectification nonlinearity, and we conjecture that this algorithm will find such nonnegative well-grounded independent sources, under reasonable initial conditions. While the algorithm has proved difficult to analyze in the general case, we give some analytical results that are consistent with this conjecture and some numerical simulations that illustrate its operation.

We consider the task of solving the independent component analysis (ICA) problem x=As given observations x, with a constraint of nonnegativity of the source random vector s. We refer to this as nonnegative independent component analysis and we consider methods for solving this task. For independent sources with nonzero probability density function (pdf) p(s) down to s=0 it is sufficient to find the orthonormal rotation y=Wz of prewhitened sources z=Vx, which minimizes the mean squared error of the reconstruction of z from the rectified version y/sup +/ of y. We suggest some algorithms which perform this, both based on a nonlinear principal component analysis (PCA) approach and on a geodesic search method driven by differential geometry considerations. We demonstrate the operation of these algorithms on an image separation problem, which shows in particular the fast convergence of the rotation and geodesic methods and apply the approach to a musical audio analysis task.

In this article, we give an overview of a range of approaches to the analysis and separation of musical audio. In particular, we consider the problems of automatic music transcription and audio source separation, which are of particular interest to our group. Monophonic music transcription, where a single note is present at one time, can be tackled using an autocorrelation-based method. For polyphonic music transcription, with several notes at any time, other approaches can be used, such as a blackboard model or a multiple-cause/sparse coding method. The latter is based on ideas and methods related to independent component analysis (ICA), a method for sound source separation.

We consider the noiseless linear independent component analysis problem, in the case where the hidden sources s are nonnegative. We assume that the random variables si are well grounded in that they have a nonvanishing probability density function (PDF) in the (positive) neighborhood of zero. For an orthonormal rotation y=Wx of prewhitened observations x=QAs, under certain reasonable conditions we show that y is a permutation of the s (apart from a scaling factor) if and only if y is nonnegative with probability 1. We suggest that this may enable the construction of practical learning algorithms, particularly for sparse nonnegative sources.

Neural networks (NNs) trained with reinforcement learning (RL) have the ability to produce complex, and robust behaviour which may be beneficial to language processing tasks. A method is proposed using RL to train NNs so that they might find synonyms that exist within a regular language. The learning algorithm and exploration strategy produces agents which yield consistently sub-optimal policies for expressions containing one operator, and unreliable performance over all expressions. This is surprising since previous work with lookup tables produced synonyms using a larger set of expressions for a wide range of learning rates and very little exploration.

It is seen that neural networks and communications are interlinked in a number of ways, towards the goal of efficient communication of information. One concrete example of this is the use of neural networks to ensure efficient use of communication channels, through connection admission control in ATM networks. In addition, however, efficient communication is also important within a decision making system such as a neural network. Finally we examine the type of neural network solutions that are suggested by this approach.

Consideration is given to image filtering (temporal and spatial) in a neural system for transmitting images through a limited capacity channel, in the case of a noisy image at the receptors. The authors use an extension of Shannon's (1949) formula for the capacity of a Gaussian channel to determine the optimum filter to be used. For realistic image statistics, they show that the bandwidth of this filter is self-limiting, and it has a high frequency boost that disappears at low signal levels. This behavior is mirrored in biological retinas.>