Search

Different methods can be employed to render virtual reverberation, often requiring substantial information about the room's geometry and the acoustic characteristics of the surfaces. However, fully comprehensive approaches that account for all aspects of a given environment may be computationally costly and redundant from a perceptual standpoint. For these methods, achieving a trade-off between perceptual authenticity and model's complexity becomes a relevant challenge. This study investigates this compromise through the use of geometrical acoustics to render Ambisonics-based binaural reverberation. Its precision is determined, among other factors, by its fidelity to the room's geometry and to the acoustic properties of its materials. The purpose of this study is to investigate the impact of simplifying the room geometry and the frequency resolution of absorption coefficients on the perception of reverberation within a virtual sound scene. Several decimated models based on a single room were perceptually evaluated using the a multi-stimulus comparison method. Additionally, these differences were numerically assessed through the calculation of acoustic parameters of the reverberation. According to numerical and perceptual evaluations, lowering the frequency resolution of absorption coefficients can have a significant impact on the perception of reverberation, while a less notable impact was observed when decimating the geometry of the model., Comment: 15 pages, 10 figures, 5 tables, will be submitted to IEEE transactions on Audio, Speech and Language processing after revisions

Vibrometry-based side channels pose a significant privacy risk, exploiting sensors like mmWave radars, light sensors, and accelerometers to detect vibrations from sound sources or proximate objects, enabling speech eavesdropping. Despite various proposed defenses, these involve costly hardware solutions with inherent physical limitations. This paper presents EveGuard, a software-driven defense framework that creates adversarial audio, protecting voice privacy from side channels without compromising human perception. We leverage the distinct sensing capabilities of side channels and traditional microphones where side channels capture vibrations and microphones record changes in air pressure, resulting in different frequency responses. EveGuard first proposes a perturbation generator model (PGM) that effectively suppresses sensor-based eavesdropping while maintaining high audio quality. Second, to enable end-to-end training of PGM, we introduce a new domain translation task called Eve-GAN for inferring an eavesdropped signal from a given audio. We further apply few-shot learning to mitigate the data collection overhead for Eve-GAN training. Our extensive experiments show that EveGuard achieves a protection rate of more than 97 percent from audio classifiers and significantly hinders eavesdropped audio reconstruction. We further validate the performance of EveGuard across three adaptive attack mechanisms. We have conducted a user study to verify the perceptual quality of our perturbed audio.

Transformers and their variants have achieved great success in speech processing. However, their multi-head self-attention mechanism is computationally expensive. Therefore, one novel selective state space model, Mamba, has been proposed as an alternative. Building on its success in automatic speech recognition, we apply Mamba for spoofing attack detection. Mamba is well-suited for this task as it can capture the artifacts in spoofed speech signals by handling long-length sequences. However, Mamba's performance may suffer when it is trained with limited labeled data. To mitigate this, we propose combining a new structure of Mamba based on a dual-column architecture with self-supervised learning, using the pre-trained wav2vec 2.0 model. The experiments show that our proposed approach achieves competitive results and faster inference on the ASVspoof 2021 LA and DF datasets, and on the more challenging In-the-Wild dataset, it emerges as the strongest candidate for spoofing attack detection. The code will be publicly released in due course., Comment: 5 pages

Recent advancements in spoken dialogue models, exemplified by systems like GPT-4o, have captured significant attention in the speech domain. Compared to traditional three-tier cascaded spoken dialogue models that comprise speech recognition (ASR), large language models (LLMs), and text-to-speech (TTS), modern spoken dialogue models exhibit greater intelligence. These advanced spoken dialogue models not only comprehend audio, music, and other speech-related features, but also capture stylistic and timbral characteristics in speech. Moreover, they generate high-quality, multi-turn speech responses with low latency, enabling real-time interaction through simultaneous listening and speaking capability. Despite the progress in spoken dialogue systems, there is a lack of comprehensive surveys that systematically organize and analyze these systems and the underlying technologies. To address this, we have first compiled existing spoken dialogue systems in the chronological order and categorized them into the cascaded and end-to-end paradigms. We then provide an in-depth overview of the core technologies in spoken dialogue models, covering aspects such as speech representation, training paradigm, streaming, duplex, and interaction capabilities. Each section discusses the limitations of these technologies and outlines considerations for future research. Additionally, we present a thorough review of relevant datasets, evaluation metrics, and benchmarks from the perspectives of training and evaluating spoken dialogue systems. We hope this survey will contribute to advancing both academic research and industrial applications in the field of spoken dialogue systems. The related material is available at https://github.com/jishengpeng/WavChat., Comment: 60 papes, working in progress

We present the MIDInfinite, a web application capable of generating symbolic music using a large-scale generative AI model locally on commodity hardware. Creating this demo involved porting the Anticipatory Music Transformer, a large language model (LLM) pre-trained on the Lakh MIDI dataset, to the Machine Learning Compilation (MLC) framework. Once the model is ported, MLC facilitates inference on a variety of runtimes including C++, mobile, and the browser. We envision that MLC has the potential to bridge the gap between the landscape of increasingly capable music AI models and technology more familiar to music software developers. As a proof of concept, we build a web application that allows users to generate endless streams of multi-instrumental MIDI in the browser, either from scratch or conditioned on a prompt. On commodity hardware (an M3 Macbook Pro), our demo can generate 51 notes per second, which is faster than real-time playback for 72.9% of generations, and increases to 86.3% with 2 seconds of upfront buffering., Comment: 2 pages

Stuttering is a neurodevelopmental speech disorder characterized by common speech symptoms such as pauses, exclamations, repetition, and prolongation. Speech-language pathologists typically assess the type and severity of stuttering by observing these symptoms. Many effective end-to-end methods exist for stuttering detection, but a commonly overlooked challenge is the uncertain relationship between tasks involved in this process. Using a suitable multi-task strategy could improve stuttering detection performance. This paper presents a novel stuttering event detection model designed to help speech-language pathologists assess both the type and severity of stuttering. First, the Conformer model extracts acoustic features from stuttered speech, followed by a Long Short-Term Memory (LSTM) network to capture contextual information. Finally, we explore multi-task learning for stuttering and propose an effective multi-task strategy. Experimental results show that our model outperforms current state-of-the-art methods for stuttering detection. In the SLT 2024 Stuttering Speech Challenge based on the AS-70 dataset [1], our model improved the mean F1 score by 24.8% compared to the baseline method and achieved first place. On this basis, we conducted relevant extensive experiments on LSTM and multi-task learning strategies respectively. The results show that our proposed method improved the mean F1 score by 39.8% compared to the baseline method., Comment: 7 pages, 3 figures, 7 tables

Computational paralinguistics (ComParal) aims to develop algorithms and models to automatically detect, analyze, and interpret non-verbal information from speech communication, e. g., emotion, health state, age, and gender. Despite its rapid progress, it heavily depends on sophisticatedly designed models given specific paralinguistic tasks. Thus, the heterogeneity and diversity of ComParal models largely prevent the realistic implementation of ComParal models. Recently, with the advent of acoustic foundation models because of self-supervised learning, developing more generic models that can efficiently perceive a plethora of paralinguistic information has become an active topic in speech processing. However, it lacks a unified evaluation framework for a fair and consistent performance comparison. To bridge this gap, we conduct a large-scale benchmark, namely ParaLBench, which concentrates on standardizing the evaluation process of diverse paralinguistic tasks, including critical aspects of affective computing such as emotion recognition and emotion dimensions prediction, over different acoustic foundation models. This benchmark contains ten datasets with thirteen distinct paralinguistic tasks, covering short-, medium- and long-term characteristics. Each task is carried out on 14 acoustic foundation models under a unified evaluation framework, which allows for an unbiased methodological comparison and offers a grounded reference for the ComParal community. Based on the insights gained from ParaLBench, we also point out potential research directions, i.e., the cross-corpus generalizability, to propel ComParal research in the future. The code associated with this study will be available to foster the transparency and replicability of this work for succeeding researchers.

Zero-shot voice conversion aims to transform a source speech utterance to match the timbre of a reference speech from an unseen speaker. Traditional approaches struggle with timbre leakage, insufficient timbre representation, and mismatches between training and inference tasks. We propose Seed-VC, a novel framework that addresses these issues by introducing an external timbre shifter during training to perturb the source speech timbre, mitigating leakage and aligning training with inference. Additionally, we employ a diffusion transformer that leverages the entire reference speech context, capturing fine-grained timbre features through in-context learning. Experiments demonstrate that Seed-VC outperforms strong baselines like OpenVoice and CosyVoice, achieving higher speaker similarity and lower word error rates in zero-shot voice conversion tasks. We further extend our approach to zero-shot singing voice conversion by incorporating fundamental frequency (F0) conditioning, resulting in comparative performance to current state-of-the-art methods. Our findings highlight the effectiveness of Seed-VC in overcoming core challenges, paving the way for more accurate and versatile voice conversion systems.

With the increasing implementation of machine learning models on edge or Internet-of-Things (IoT) devices, deploying advanced models on resource-constrained IoT devices remains challenging. Transformer models, a currently dominant neural architecture, have achieved great success in broad domains but their complexity hinders its deployment on IoT devices with limited computation capability and storage size. Although many model compression approaches have been explored, they often suffer from notorious performance degradation. To address this issue, we introduce a new method, namely Transformer Re-parameterization, to boost the performance of lightweight Transformer models. It consists of two processes: the High-Rank Factorization (HRF) process in the training stage and the deHigh-Rank Factorization (deHRF) process in the inference stage. In the former process, we insert an additional linear layer before the Feed-Forward Network (FFN) of the lightweight Transformer. It is supposed that the inserted HRF layers can enhance the model learning capability. In the later process, the auxiliary HRF layer will be merged together with the following FFN layer into one linear layer and thus recover the original structure of the lightweight model. To examine the effectiveness of the proposed method, we evaluate it on three widely used Transformer variants, i.e., ConvTransformer, Conformer, and SpeechFormer networks, in the application of speech emotion recognition on the IEMOCAP, M3ED and DAIC-WOZ datasets. Experimental results show that our proposed method consistently improves the performance of lightweight Transformers, even making them comparable to large models. The proposed re-parameterization approach enables advanced Transformer models to be deployed on resource-constrained IoT devices.

In this study, we propose an ensemble learning framework for electroencephalogram-based overt speech classification, leveraging denoising diffusion probabilistic models with varying convolutional kernel sizes. The ensemble comprises three models with kernel sizes of 51, 101, and 201, effectively capturing multi-scale temporal features inherent in signals. This approach improves the robustness and accuracy of speech decoding by accommodating the rich temporal complexity of neural signals. The ensemble models work in conjunction with conditional autoencoders that refine the reconstructed signals and maximize the useful information for downstream classification tasks. The results indicate that the proposed ensemble-based approach significantly outperforms individual models and existing state-of-the-art techniques. These findings demonstrate the potential of ensemble methods in advancing brain signal decoding, offering new possibilities for non-verbal communication applications, particularly in brain-computer interface systems aimed at aiding individuals with speech impairments.

Brain signals accompany various information relevant to human actions and mental imagery, making them crucial to interpreting and understanding human intentions. Brain-computer interface technology leverages this brain activity to generate external commands for controlling the environment, offering critical advantages to individuals with paralysis or locked-in syndrome. Within the brain-computer interface domain, brain-to-speech research has gained attention, focusing on the direct synthesis of audible speech from brain signals. Most current studies decode speech from brain activity using invasive techniques and emphasize spoken speech data. However, humans express various speech states, and distinguishing these states through non-invasive approaches remains a significant yet challenging task. This research investigated the effectiveness of deep learning models for non-invasive-based neural signal decoding, with an emphasis on distinguishing between different speech paradigms, including perceived, overt, whispered, and imagined speech, across multiple frequency bands. The model utilizing the spatial conventional neural network module demonstrated superior performance compared to other models, especially in the gamma band. Additionally, imagined speech in the theta frequency band, where deep learning also showed strong effects, exhibited statistically significant differences compared to the other speech paradigms.

Given the extensive research and real-world applications of automatic speech recognition (ASR), ensuring the robustness of ASR models against minor input perturbations becomes a crucial consideration for maintaining their effectiveness in real-time scenarios. Previous explorations into ASR model robustness have predominantly revolved around evaluating accuracy on white-box settings with full access to ASR models. Nevertheless, full ASR model details are often not available in real-world applications. Therefore, evaluating the robustness of black-box ASR models is essential for a comprehensive understanding of ASR model resilience. In this regard, we thoroughly study the vulnerability of practical black-box attacks in cutting-edge ASR models and propose to employ two advanced time-domain-based transferable attacks alongside our differentiable feature extractor. We also propose a speech-aware gradient optimization approach (SAGO) for ASR, which forces mistranscription with minimal impact on human imperceptibility through voice activity detection rule and a speech-aware gradient-oriented optimizer. Our comprehensive experimental results reveal performance enhancements compared to baseline approaches across five models on two databases., Comment: IEEE SPL

This paper builds upon an existing speech emotion recognition model by adding an additional LSTM layer to improve the accuracy and processing efficiency of emotion recognition from audio data. By capturing the long-term dependencies within audio sequences through a dual-layer LSTM network, the model can recognize and classify complex emotional patterns more accurately. Experiments conducted on the RAVDESS dataset validated this approach, showing that the modified dual layer LSTM model improves accuracy by 2% compared to the single-layer LSTM while significantly reducing recognition latency, thereby enhancing real-time performance. These results indicate that the dual-layer LSTM architecture is highly suitable for handling emotional features with long-term dependencies, providing a viable optimization for speech emotion recognition systems. This research provides a reference for practical applications in fields like intelligent customer service, sentiment analysis and human-computer interaction.

Speech-to-Speech Translation (S2ST) models transform speech from one language to another target language with the same linguistic information. S2ST is important for bridging the communication gap among communities and has diverse applications. In recent years, researchers have introduced direct S2ST models, which have the potential to translate speech without relying on intermediate text generation, have better decoding latency, and the ability to preserve paralinguistic and non-linguistic features. However, direct S2ST has yet to achieve quality performance for seamless communication and still lags behind the cascade models in terms of performance, especially in real-world translation. To the best of our knowledge, no comprehensive survey is available on the direct S2ST system, which beginners and advanced researchers can look upon for a quick survey. The present work provides a comprehensive review of direct S2ST models, data and application issues, and performance metrics. We critically analyze the models' performance over the benchmark datasets and provide research challenges and future directions.

With the rise of Speech Large Language Models (Speech LLMs), there has been growing interest in discrete speech tokens for their ability to integrate with text-based tokens seamlessly. Compared to most studies that focus on continuous speech features, although discrete-token based LLMs have shown promising results on certain tasks, the performance gap between these two paradigms is rarely explored. In this paper, we present a fair and thorough comparison between discrete and continuous features across a variety of semantic-related tasks using a light-weight LLM (Qwen1.5-0.5B). Our findings reveal that continuous features generally outperform discrete tokens, particularly in tasks requiring fine-grained semantic understanding. Moreover, this study goes beyond surface-level comparison by identifying key factors behind the under-performance of discrete tokens, such as limited token granularity and inefficient information retention. To enhance the performance of discrete tokens, we explore potential aspects based on our analysis. We hope our results can offer new insights into the opportunities for advancing discrete speech tokens in Speech LLMs., Comment: 5 tables, 4 figures

The estimation of room impulse responses (RIRs) between static loudspeaker and microphone locations can be done using a number of well-established measurement and inference procedures. While these procedures assume a time-invariant acoustic system, time variations need to be considered for the case of spatially dynamic scenarios where loudspeakers and microphones are subject to movement. If the RIR is modeled using image sources, then movement implies that the distance to each image source varies over time, making the estimation of the spatially dynamic RIR particularly challenging. In this paper, we propose a procedure to estimate the early part of the spatially dynamic RIR between a stationary source and a microphone moving on a linear trajectory at constant velocity. The procedure is built upon a state-space model, where the state to be estimated represents the early RIR, the observation corresponds to a microphone recording in a spatially dynamic scenario, and time-varying distances to the image sources are incorporated into the state transition matrix obtained from static RIRs at the start and end point of the trajectory. The performance of the proposed approach is evaluated against state-of-the-art RIR interpolation and state-space estimation methods using simulations, demonstrating the potential of the proposed state-space model., Comment: 30 pages, 13 figures

This paper addresses the challenge of speaker separation, which remains an active research topic despite the promising results achieved in recent years. These results, however, often degrade in real recording conditions due to the presence of noise, echo, and other interferences. This is because neural models are typically trained on synthetic datasets consisting of mixed audio signals and their corresponding ground truths, which are generated using computer software and do not fully represent the complexities of real-world recording scenarios. The lack of realistic training sets for speaker separation remains a major hurdle, as obtaining individual sounds from mixed audio signals is a nontrivial task. To address this issue, we propose a novel method for constructing a realistic training set that includes mixture signals and corresponding ground truths for each speaker. We evaluate this dataset on a deep learning model and compare it to a synthetic dataset. We got a 1.65 dB improvement in Scale Invariant Signal to Distortion Ratio (SI-SDR) for speaker separation accuracy in realistic mixing. Our findings highlight the potential of realistic training sets for enhancing the performance of speaker separation models in real-world scenarios., Comment: in Arabic language

AI-synthesized speech, also known as deepfake speech, has recently raised significant concerns due to the rapid advancement of speech synthesis and speech conversion techniques. Previous works often rely on distinguishing synthesizer artifacts to identify deepfake speech. However, excessive reliance on these specific synthesizer artifacts may result in unsatisfactory performance when addressing speech signals created by unseen synthesizers. In this paper, we propose a robust deepfake speech detection method that employs feature decomposition to learn synthesizer-independent content features as complementary for detection. Specifically, we propose a dual-stream feature decomposition learning strategy that decomposes the learned speech representation using a synthesizer stream and a content stream. The synthesizer stream specializes in learning synthesizer features through supervised training with synthesizer labels. Meanwhile, the content stream focuses on learning synthesizer-independent content features, enabled by a pseudo-labeling-based supervised learning method. This method randomly transforms speech to generate speed and compression labels for training. Additionally, we employ an adversarial learning technique to reduce the synthesizer-related components in the content stream. The final classification is determined by concatenating the synthesizer and content features. To enhance the model's robustness to different synthesizer characteristics, we further propose a synthesizer feature augmentation strategy that randomly blends the characteristic styles within real and fake audio features and randomly shuffles the synthesizer features with the content features. This strategy effectively enhances the feature diversity and simulates more feature combinations.

Music therapy has been shown in recent years to provide multiple health benefits related to emotional wellness. In turn, maintaining a healthy emotional state has proven to be effective for patients undergoing treatment, such as Parkinson's patients or patients suffering from stress and anxiety. We propose fine-tuning MusicGen, a music-generating transformer model, to create short musical clips that assist patients in transitioning from negative to desired emotional states. Using low-rank decomposition fine-tuning on the MTG-Jamendo Dataset with emotion tags, we generate 30-second clips that adhere to the iso principle, guiding patients through intermediate states in the valence-arousal circumplex. The generated music is evaluated using a music emotion recognition model to ensure alignment with intended emotions. By concatenating these clips, we produce a 15-minute "music medicine" resembling a music therapy session. Our approach is the first model to leverage Language Models to generate music medicine. Ultimately, the output is intended to be used as a temporary relief between music therapy sessions with a board-certified therapist., Comment: Late-Breaking / Demo Session Extended Abstract, ISMIR 2024 Conference

Album sequencing is a critical part of the album production process. Recently, a data-driven approach was proposed that sequences general collections of independent media by extracting the narrative essence of the items in the collections. While this approach implies an album sequencing technique, it is not widely accessible to a less technical audience, requiring advanced knowledge of machine learning techniques to use. To address this, we introduce a new user-friendly web-based tool that allows a less technical audience to upload music tracks, execute this technique in one click, and subsequently presents the result in a clean visualization to the user. To both increase the number of templates available to the user and address shortcomings of previous work, we also introduce a new direct transformer-based album sequencing method. We find that our more direct method outperforms a random baseline but does not reach the same performance as the narrative essence approach. Both methods are included in our web-based user interface, and this -- alongside a full copy of our implementation -- is publicly available at https://github.com/dylanashley/automatic-album-sequencing, Comment: presented as a late breaking demo in the 25th International Society for Music Information Retrieval Conference; 3 pages in main text + 1 page of references, 3 figures in main text; source code available at https://github.com/dylanashley/automatic-album-sequencing

Optimization of a trade-off between the number of speakers and their temporal variability (or session diversity) is crucial for the development of a speaker recognition system together with making the data collection process feasible from a time perspective. In this article, we provide the analysis of dependency between inter and intra speaker variability in training data for the modern neural network-based speaker recognition system using the VoxTube dataset for text-independent speaker recognition task. Besides, an auxiliary contribution of this work is a release of upload date metadata per utterance in a VoxTube dataset. We want this article to contribute to guidelines and best practices for collecting and filtering data from media hosting platforms to facilitate the efforts of researchers in developing speaker recognition systems.

Speech enhancement plays an essential role in various applications, and the integration of visual information has been demonstrated to bring substantial advantages. However, the majority of current research concentrates on the examination of facial and lip movements, which can be compromised or entirely inaccessible in scenarios where occlusions occur or when the camera view is distant. Whereas contextual visual cues from the surrounding environment have been overlooked: for example, when we see a dog bark, our brain has the innate ability to discern and filter out the barking noise. To this end, in this paper, we introduce a novel task, i.e. SAV-SE. To our best knowledge, this is the first proposal to use rich contextual information from synchronized video as auxiliary cues to indicate the type of noise, which eventually improves the speech enhancement performance. Specifically, we propose the VC-S$^2$E method, which incorporates the Conformer and Mamba modules for their complementary strengths. Extensive experiments are conducted on public MUSIC, AVSpeech and AudioSet datasets, where the results demonstrate the superiority of VC-S$^2$E over other competitive methods. We will make the source code publicly available. Project demo page: https://AVSEPage.github.io/

CTC compressor can be an effective approach to integrate audio encoders to decoder-only models, which has gained growing interest for different speech applications. In this work, we propose a novel CTC compressor based joint speech and text training (CJST) framework for decoder-only ASR. CJST matches speech and text modalities from both directions by exploring a simple modality adaptor and several features of the CTC compressor, including sequence compression, on-the-fly forced peaky alignment and CTC class embeddings. Experimental results on the Librispeech and TED-LIUM2 corpora show that the proposed CJST achieves an effective text injection without the need of duration handling, leading to the best performance for both in-domain and cross-domain scenarios. We also provide a comprehensive study on CTC compressor, covering various compression modes, edge case handling and behavior under both clean and noisy data conditions, which reveals the most robust setting to use CTC compressor for decoder-only models., Comment: submitted to ICASSP2025

Recent advances in voice synthesis, coupled with the ease with which speech can be harvested for millions of people, introduce new threats to applications that are enabled by devices such as voice assistants (e.g., Amazon Alexa, Google Home etc.). We explore if unrelated and limited amount of speech from a target can be used to synthesize commands for a voice assistant like Amazon Alexa. More specifically, we investigate attacks on voice assistants with synthetic commands when they match command sources to authorized users, and applications (e.g., Alexa Skills) process commands only when their source is an authorized user with a chosen confidence level. We demonstrate that even simple concatenative speech synthesis can be used by an attacker to command voice assistants to perform sensitive operations. We also show that such attacks, when launched by exploiting compromised devices in the vicinity of voice assistants, can have relatively small host and network footprint. Our results demonstrate the need for better defenses against synthetic malicious commands that could target voice assistants.

Music generation has progressed significantly, especially in the domain of audio generation. However, generating symbolic music that is both long-structured and expressive remains a significant challenge. In this paper, we propose PerceiverS (Segmentation and Scale), a novel architecture designed to address this issue by leveraging both Effective Segmentation and Multi-Scale attention mechanisms. Our approach enhances symbolic music generation by simultaneously learning long-term structural dependencies and short-term expressive details. By combining cross-attention and self-attention in a Multi-Scale setting, PerceiverS captures long-range musical structure while preserving performance nuances. The proposed model, evaluated on datasets like Maestro, demonstrates improvements in generating coherent and diverse music with both structural consistency and expressive variation. The project demos and the generated music samples can be accessed through the link: https://perceivers.github.io.

Text toxicity detection systems exhibit significant biases, producing disproportionate rates of false positives on samples mentioning demographic groups. But what about toxicity detection in speech? To investigate the extent to which text-based biases are mitigated by speech-based systems, we produce a set of high-quality group annotations for the multilingual MuTox dataset, and then leverage these annotations to systematically compare speech- and text-based toxicity classifiers. Our findings indicate that access to speech data during inference supports reduced bias against group mentions, particularly for ambiguous and disagreement-inducing samples. Our results also suggest that improving classifiers, rather than transcription pipelines, is more helpful for reducing group bias. We publicly release our annotations and provide recommendations for future toxicity dataset construction.

Speech impairments in Parkinson's disease (PD) provide significant early indicators for diagnosis. While models for speech-based PD detection have shown strong performance, their interpretability remains underexplored. This study systematically evaluates several explainability methods to identify PD-specific speech features, aiming to support the development of accurate, interpretable models for clinical decision-making in PD diagnosis and monitoring. Our methodology involves (i) obtaining attributions and saliency maps using mainstream interpretability techniques, (ii) quantitatively evaluating the faithfulness of these maps and their combinations obtained via union and intersection through a range of established metrics, and (iii) assessing the information conveyed by the saliency maps for PD detection from an auxiliary classifier. Our results reveal that, while explanations are aligned with the classifier, they often fail to provide valuable information for domain experts., Comment: The first two authors contributed equally to this research: author order is alphabetical

Neural vocoders often struggle with aliasing in latent feature spaces, caused by time-domain nonlinear operations and resampling layers. Aliasing folds high-frequency components into the low-frequency range, making aliased and original frequency components indistinguishable and introducing two practical issues. First, aliasing complicates the waveform generation process, as the subsequent layers must address these aliasing effects, increasing the computational complexity. Second, it limits extrapolation performance, particularly in handling high fundamental frequencies, which degrades the perceptual quality of generated speech waveforms. This paper demonstrates that 1) time-domain nonlinear operations inevitably introduce aliasing but provide a strong inductive bias for harmonic generation, and 2) time-frequency-domain processing can achieve aliasing-free waveform synthesis but lacks the inductive bias for effective harmonic generation. Building on this insight, we propose Wavehax, an aliasing-free neural WAVEform generator that integrates 2D convolution and a HArmonic prior for reliable Complex Spectrogram estimation. Experimental results show that Wavehax achieves speech quality comparable to existing high-fidelity neural vocoders and exhibits exceptional robustness in scenarios requiring high fundamental frequency extrapolation, where aliasing effects become typically severe. Moreover, Wavehax requires less than 5% of the multiply-accumulate operations and model parameters compared to HiFi-GAN V1, while achieving over four times faster CPU inference speed., Comment: 13 pages, 5 figures, Submitted to IEEE/ACM Trans. ASLP

Auscultation of internal body sounds is essential for diagnosing a range of health conditions, yet its effectiveness is often limited by clinicians' expertise and the acoustic constraints of human hearing, restricting its use across various clinical scenarios. To address these challenges, we introduce AuscultaBase, a foundational framework aimed at advancing body sound diagnostics through innovative data integration and contrastive learning techniques. Our contributions include the following: First, we compile AuscultaBase-Corpus, a large-scale, multi-source body sound database encompassing 11 datasets with 40,317 audio recordings and totaling 322.4 hours of heart, lung, and bowel sounds. Second, we develop AuscultaBase-Model, a foundational diagnostic model for body sounds, utilizing contrastive learning on the compiled corpus. Third, we establish AuscultaBase-Bench, a comprehensive benchmark containing 16 sub-tasks, assessing the performance of various open-source acoustic pre-trained models. Evaluation results indicate that our model outperforms all other open-source models in 12 out of 16 tasks, demonstrating the efficacy of our approach in advancing diagnostic capabilities for body sound analysis., Comment: 26 pages

Development of optical technology has enabled imaging of two-dimensional (2D) sound fields. This acousto-optic sensing enables understanding of the interaction between sound and objects such as reflection and diffraction. Moreover, it is expected to be used an advanced measurement technology for sonars in self-driving vehicles and assistive robots. However, the low sound-pressure sensitivity of the acousto-optic sensing results in high intensity of noise on images. Therefore, denoising is an essential task to visualize and analyze the sound fields. In addition to denoising, segmentation of sound and object silhouette is also required to analyze interactions between them. In this paper, we propose sound-field-images-with-object-silhouette denoising and segmentation (SoundSil-DS) that jointly perform denoising and segmentation for sound fields and object silhouettes on a visualized image. We developed a new model based on the current state-of-the-art denoising network. We also created a dataset to train and evaluate the proposed method through acoustic simulation. The proposed method was evaluated using both simulated and measured data. We confirmed that our method can applied to experimentally measured data. These results suggest that the proposed method may improve the post-processing for sound fields, such as physical model-based three-dimensional reconstruction since it can remove unwanted noise and separate sound fields and other object silhouettes. Our code is available at https://github.com/nttcslab/soundsil-ds., Comment: 13 pages, 12 figures, 5 tables. Accepted by WACV 2025

A conversational music retrieval system can help users discover music that matches their preferences through dialogue. To achieve this, a conversational music retrieval system should seamlessly engage in multi-turn conversation by 1) understanding user queries and 2) responding with natural language and retrieved music. A straightforward solution would be a data-driven approach utilizing such conversation logs. However, few datasets are available for the research and are limited in terms of volume and quality. In this paper, we present a data generation framework for rich music discovery dialogue using a large language model (LLM) and user intents, system actions, and musical attributes. This is done by i) dialogue intent analysis using grounded theory, ii) generating attribute sequences via cascading database filtering, and iii) generating utterances using large language models. By applying this framework to the Million Song dataset, we create LP-MusicDialog, a Large Language Model based Pseudo Music Dialogue dataset, containing over 288k music conversations using more than 319k music items. Our evaluation shows that the synthetic dataset is competitive with an existing, small human dialogue dataset in terms of dialogue consistency, item relevance, and naturalness. Furthermore, using the dataset, we train a conversational music retrieval model and show promising results., Comment: Accepted for publication at the 25th International Society for Music Information Retrieval Conference (ISMIR 2024)

We propose an efficient workflow for high-quality offline alignment of in-the-wild performance audio and corresponding sheet music scans (images). Recent work on audio-to-score alignment extends dynamic time warping (DTW) to be theoretically able to handle jumps in sheet music induced by repeat signs-this method requires no human annotations, but we show that it often yields low-quality alignments. As an alternative, we propose a workflow and interface that allows users to quickly annotate jumps (by clicking on repeat signs), requiring a small amount of human supervision but yielding much higher quality alignments on average. Additionally, we refine audio and score feature representations to improve alignment quality by: (1) integrating measure detection into the score feature representation, and (2) using raw onset prediction probabilities from a music transcription model instead of piano roll. We propose an evaluation protocol for audio-to-score alignment that computes the distance between the estimated and ground truth alignment in units of measures. Under this evaluation, we find that our proposed jump annotation workflow and improved feature representations together improve alignment accuracy by 150% relative to prior work (33% to 82%)., Comment: 25th International Society for Music Information Retrieval Conference, San Francisco, 2024

End-to-end speech translation (ST), which translates source language speech directly into target language text, has garnered significant attention in recent years. Many ST applications require strict length control to ensure that the translation duration matches the length of the source audio, including both speech and pause segments. Previous methods often controlled the number of words or characters generated by the Machine Translation model to approximate the source sentence's length without considering the isochrony of pauses and speech segments, as duration can vary between languages. To address this, we present improvements to the duration alignment component of our sequence-to-sequence ST model. Our method controls translation length by predicting the duration of speech and pauses in conjunction with the translation process. This is achieved by providing timing information to the decoder, ensuring it tracks the remaining duration for speech and pauses while generating the translation. The evaluation on the Zh-En test set of CoVoST 2, demonstrates that the proposed Isochrony-Controlled ST achieves 0.92 speech overlap and 8.9 BLEU, which has only a 1.4 BLEU drop compared to the ST baseline.

Audio super-resolution aims to enhance low-resolution signals by creating high-frequency content. In this work, we modify the architecture of AERO (a state-of-the-art system for this task) for music super-resolution. SPecifically, we replace its original Attention and LSTM layers with Mamba, a State Space Model (SSM), across all network layers. Mamba is capable of effectively substituting the mentioned modules, as it offers a mechanism similar to that of Attention while also functioning as a recurrent network. With the proposed AEROMamba, training requires 2-4x less GPU memory, since Mamba exploits the convolutional formulation and leverages GPU memory hierarchy. Additionally, during inference, Mamba operates in constant memory due to recurrence, avoiding memory growth associated with Attention. This results in a 14x speed improvement using 5x less GPU. Subjective listening tests (0 to 100 scale) show that the proposed model surpasses the AERO model. In the MUSDB dataset, degraded signals scored 38.22, while AERO and AEROMamba scored 60.03 and 66.74, respectively. For the PianoEval dataset, scores were 72.92 for degraded signals, 76.89 for AERO, and 84.41 for AEROMamba., Comment: Accepted at LAMIR 2024 Workshop (ISMIR 2024 Satellite Event)

Large language models (LLMs) prompted with text and audio represent the state of the art in various auditory tasks, including speech, music, and general audio, showing emergent abilities on unseen tasks. However, these capabilities have yet to be fully demonstrated in bioacoustics tasks, such as detecting animal vocalizations in large recordings, classifying rare and endangered species, and labeling context and behavior - tasks that are crucial for conservation, biodiversity monitoring, and the study of animal behavior. In this work, we present NatureLM-audio, the first audio-language foundation model specifically designed for bioacoustics. Our carefully curated training dataset comprises text-audio pairs spanning a diverse range of bioacoustics, speech, and music data, designed to address the challenges posed by limited annotated datasets in the field. We demonstrate successful transfer of learned representations from music and speech to bioacoustics, and our model shows promising generalization to unseen taxa and tasks. Importantly, we test NatureLM-audio on a novel benchmark (BEANS-Zero) and it sets the new state of the art (SotA) on several bioacoustics tasks, including zero-shot classification of unseen species. To advance bioacoustics research, we also open-source the code for generating training and benchmark data, as well as for training the model., Comment: Demo page: https://earthspecies.github.io/naturelm-audio-demo/ The code will be open-sourced and available shortly

This technical report presents our initial attempt to build a spoken large language model (LLM) for Taiwanese Mandarin, specifically tailored to enable real-time, speech-to-speech interaction in multi-turn conversations. Our end-to-end model incorporates a decoder-only transformer architecture and aims to achieve seamless interaction while preserving the conversational flow, including full-duplex capabilities allowing simultaneous speaking and listening. The paper also details the training process, including data preparation with synthesized dialogues and adjustments for real-time interaction. We also developed a platform to evaluate conversational fluency and response coherence in multi-turn dialogues. We hope the release of the report can contribute to the future development of spoken LLMs in Taiwanese Mandarin., Comment: Work in progress

Selective state space models (SSMs) represented by Mamba have demonstrated their computational efficiency and promising outcomes in various tasks, including automatic speech recognition (ASR). Mamba has been applied to ASR task with the attention-based encoder-decoder framework, where the cross-attention mechanism between encoder and decoder remains. This paper explores the capability of Mamba as the decoder-only architecture in ASR task. Our MAmba-based DEcoder-ONly approach (MADEON) consists of a single decoder that takes speech tokens as a condition and predicts text tokens in an autoregressive manner. To enhance MADEON, we further propose speech prefixing that performs bidirectional processing on speech tokens, which enriches the contextual information in the hidden states. Our experiments show that MADEON significantly outperforms a non-selective SSM. The combination of speech prefixing and the recently proposed Mamba-2 yields comparable performance to Transformer-based models on large datasets., Comment: Accepted to SLT 2024

Decoding the directional focus of an attended speaker from listeners' electroencephalogram (EEG) signals is essential for developing brain-computer interfaces to improve the quality of life for individuals with hearing impairment. Previous works have concentrated on binary directional focus decoding, i.e., determining whether the attended speaker is on the left or right side of the listener. However, a more precise decoding of the exact direction of the attended speaker is necessary for effective speech processing. Additionally, audio spatial information has not been effectively leveraged, resulting in suboptimal decoding results. In this paper, we observe that, on our recently presented dataset with 15-class directional focus, models relying exclusively on EEG inputs exhibits significantly lower accuracy when decoding the directional focus in both leave-one-subject-out and leave-one-trial-out scenarios. By integrating audio spatial spectra with EEG features, the decoding accuracy can be effectively improved. We employ the CNN, LSM-CNN, and EEG-Deformer models to decode the directional focus from listeners' EEG signals with the auxiliary audio spatial spectra. The proposed Sp-Aux-Deformer model achieves notable 15-class decoding accuracies of 57.48% and 61.83% in leave-one-subject-out and leave-one-trial-out scenarios, respectively.

We propose a single-channel Deep Cascade Fusion of Diarization and Separation (DCF-DS) framework for back-end speech recognition, combining neural speaker diarization (NSD) and speech separation (SS). First, we sequentially integrate the NSD and SS modules within a joint training framework, enabling the separation module to leverage speaker time boundaries from the diarization module effectively. Then, to complement DCF-DS training, we introduce a window-level decoding scheme that allows the DCF-DS framework to handle the sparse data convergence instability (SDCI) problem. We also explore using an NSD system trained on real datasets to provide more accurate speaker boundaries during decoding. Additionally, we incorporate an optional multi-input multi-output speech enhancement module (MIMO-SE) within the DCF-DS framework, which offers further performance gains. Finally, we enhance diarization results by re-clustering DCF-DS outputs, improving ASR accuracy. By incorporating the DCF-DS method, we achieved first place in the realistic single-channel track of the CHiME-8 NOTSOFAR-1 challenge. We also perform the evaluation on the open LibriCSS dataset, achieving a new state-of-the-art single-channel speech recognition performance.

In our demo, participants are invited to explore the Diff-MSTC prototype, which integrates the Diff-MST model into Steinberg's digital audio workstation (DAW), Cubase. Diff-MST, a deep learning model for mixing style transfer, forecasts mixing console parameters for tracks using a reference song. The system processes up to 20 raw tracks along with a reference song to predict mixing console parameters that can be used to create an initial mix. Users have the option to manually adjust these parameters further for greater control. In contrast to earlier deep learning systems that are limited to research ideas, Diff-MSTC is a first-of-its-kind prototype integrated into a DAW. This integration facilitates mixing decisions on multitracks and lets users input context through a reference song, followed by fine-tuning of audio effects in a traditional manner., Comment: Presented at 2024 International Society for Music Information Retrieval

In debating, rebuttal is one of the most critical stages, where a speaker addresses the arguments presented by the opposing side. During this process, the speaker synthesizes their own persuasive articulation given the context from the opposing side. This work proposes a novel zero-shot text-to-speech synthesis system for rebuttal, namely Debatts. Debatts takes two speech prompts, one from the opposing side (i.e. opponent) and one from the speaker. The prompt from the opponent is supposed to provide debating style prosody, and the prompt from the speaker provides identity information. In particular, we pretrain the Debatts system from in-the-wild dataset, and integrate an additional reference encoder to take debating prompt for style. In addition, we also create a debating dataset to develop Debatts. In this setting, Debatts can generate a debating-style speech in rebuttal for any voices. Experimental results confirm the effectiveness of the proposed system in comparison with the classic zero-shot TTS systems.

Contextual ASR or hotword customization holds substantial practical value. Despite the impressive performance of current end-to-end (E2E) automatic speech recognition (ASR) systems, they often face challenges in accurately recognizing rare words. Typical E2E contextual ASR models commonly feature complex architectures and decoding mechanisms, limited in performance and susceptible to interference from distractor words. With large language model (LLM)-based ASR models emerging as the new mainstream, we propose a CTC-Assisted LLM-Based Contextual ASR model with an efficient filtering algorithm. By using coarse CTC decoding results to filter potential relevant hotwords and incorporating them into LLM prompt input, our model attains WER/B-WER of 1.27%/3.67% and 2.72%/8.02% on the Librispeech test-clean and test-other sets targeting on recognizing rare long-tail words, demonstrating significant improvements compared to the baseline LLM-based ASR model, and substantially surpassing other related work. More remarkably, with the help of the large language model and proposed filtering algorithm, our contextual ASR model still performs well with 2000 biasing words., Comment: SLT 2024

Sound event localization and detection (SELD) has seen substantial advancements through learning-based methods. These systems, typically trained from scratch on specific datasets, have shown considerable generalization capabilities. Recently, deep neural networks trained on large-scale datasets have achieved remarkable success in the sound event classification (SEC) field, prompting an open question of whether these advancements can be extended to develop general-purpose SELD models. In this paper, leveraging the power of pre-trained SEC models, we propose pre-trained SELD networks (PSELDNets) on large-scale synthetic datasets. These synthetic datasets, generated by convolving sound events with simulated spatial room impulse responses (SRIRs), contain 1,167 hours of audio clips with an ontology of 170 sound classes. These PSELDNets are transferred to downstream SELD tasks. When we adapt PSELDNets to specific scenarios, particularly in low-resource data cases, we introduce a data-efficient fine-tuning method, AdapterBit. PSELDNets are evaluated on a synthetic-test-set using collected SRIRs from TAU Spatial Room Impulse Response Database (TAU-SRIR DB) and achieve satisfactory performance. We also conduct our experiments to validate the transferability of PSELDNets to three publicly available datasets and our own collected audio recordings. Results demonstrate that PSELDNets surpass state-of-the-art systems across all publicly available datasets. Given the need for direction-of-arrival estimation, SELD generally relies on sufficient multi-channel audio clips. However, incorporating the AdapterBit, PSELDNets show more efficient adaptability to various tasks using minimal multi-channel or even just monophonic audio clips, outperforming the traditional fine-tuning approaches., Comment: 13 pages, 9 figures. The code is available at https://github.com/Jinbo-Hu/PSELDNets

Realistic audio synthesis that captures accurate acoustic phenomena is essential for creating immersive experiences in virtual and augmented reality. Synthesizing the sound received at any position relies on the estimation of impulse response (IR), which characterizes how sound propagates in one scene along different paths before arriving at the listener's position. In this paper, we present Acoustic Volume Rendering (AVR), a novel approach that adapts volume rendering techniques to model acoustic impulse responses. While volume rendering has been successful in modeling radiance fields for images and neural scene representations, IRs present unique challenges as time-series signals. To address these challenges, we introduce frequency-domain volume rendering and use spherical integration to fit the IR measurements. Our method constructs an impulse response field that inherently encodes wave propagation principles and achieves state-of-the-art performance in synthesizing impulse responses for novel poses. Experiments show that AVR surpasses current leading methods by a substantial margin. Additionally, we develop an acoustic simulation platform, AcoustiX, which provides more accurate and realistic IR simulations than existing simulators. Code for AVR and AcoustiX are available at https://zitonglan.github.io/avr., Comment: NeurIPS 2024 Spotlight

This study focuses on Intelligent Fault Diagnosis (IFD) in rotating machinery utilizing a single microphone and a data-driven methodology, effectively diagnosing 42 classes of fault types and severities. The research leverages sound data from the imbalanced MaFaulDa dataset, aiming to strike a balance between high performance and low resource consumption. The testing phase encompassed a variety of configurations, including sampling, quantization, signal normalization, silence removal, Wiener filtering, data scaling, windowing, augmentation, and classifier tuning using XGBoost. Through the analysis of time, frequency, mel-frequency, and statistical features, we achieved an impressive accuracy of 99.54% and an F-Beta score of 99.52% with just 6 boosting trees at an 8 kHz, 8-bit configuration. Moreover, when utilizing only MFCCs along with their first- and second-order deltas, we recorded an accuracy of 97.83% and an F-Beta score of 97.67%. Lastly, by implementing a greedy wrapper approach, we obtained a remarkable accuracy of 96.82% and an F-Beta score of 98.86% using 50 selected features, nearly all of which were first- and second-order deltas of the MFCCs.

Voice user interfaces (VUIs) have facilitated the efficient interactions between humans and machines through spoken commands. Since real-word acoustic scenes are complex, speech enhancement plays a critical role for robust VUI. Transformer and its variants, such as Conformer, have demonstrated cutting-edge results in speech enhancement. However, both of them suffers from the quadratic computational complexity with respect to the sequence length, which hampers their ability to handle long sequences. Recently a novel State Space Model called Mamba, which shows strong capability to handle long sequences with linear complexity, offers a solution to address this challenge. In this paper, we propose a novel hybrid convolution-Mamba backbone, denoted as MambaDC, for speech enhancement. Our MambaDC marries the benefits of convolutional networks to model the local interactions and Mamba's ability for modeling long-range global dependencies. We conduct comprehensive experiments within both basic and state-of-the-art (SoTA) speech enhancement frameworks, on two commonly used training targets. The results demonstrate that MambaDC outperforms Transformer, Conformer, and the standard Mamba across all training targets. Built upon the current advanced framework, the use of MambaDC backbone showcases superior results compared to existing \textcolor{black}{SoTA} systems. This sets the stage for efficient long-range global modeling in speech enhancement., Comment: Submitted to IEEE TCE

Speech-based assessment of the schizophrenia spectrum has been widely researched over in the recent past. In this study, we develop a deep learning framework to estimate schizophrenia severity scores from speech using a feature fusion approach that fuses articulatory features with different self-supervised speech features extracted from pre-trained audio models. We also propose an auto-encoder-based self-supervised representation learning framework to extract compact articulatory embeddings from speech. Our top-performing speech-based fusion model with Multi-Head Attention (MHA) reduces Mean Absolute Error (MAE) by 9.18% and Root Mean Squared Error (RMSE) by 9.36% for schizophrenia severity estimation when compared with the previous models that combined speech and video inputs., Comment: Submitted to ICASSP-SPADE workshop 2025

The synchronization of motor responses to rhythmic auditory cues is a fundamental biological phenomenon observed across various species. While the importance of temporal alignment varies across different contexts, achieving precise temporal synchronization is a prominent goal in musical performances. Musicians often incorporate expressive timing variations, which require precise control over timing and synchronization, particularly in ensemble performance. This is crucial because both deliberate expressive nuances and accidental timing deviations can affect the overall timing of a performance. This discussion prompts the question of how musicians adjust their temporal dynamics to achieve synchronization within an ensemble. This paper introduces a novel feedback correction model based on the Kalman Filter, aimed at improving the understanding of interpersonal timing in ensemble music performances. The proposed model performs similarly to other linear correction models in the literature, with the advantage of low computational cost and good performance even in scenarios where the underlying tempo varies., Comment: 7 pages, 1 figure. Accepted for publication on the 25th International Society for Music Information Retrieval (ISMIR 2024)

This perspective calls for scholars across disciplines to address the challenge of audio deepfake detection and discernment through an interdisciplinary lens across Artificial Intelligence methods and linguistics. With an avalanche of tools for the generation of realistic-sounding fake speech on one side, the detection of deepfakes is lagging on the other. Particularly hindering audio deepfake detection is the fact that current AI models lack a full understanding of the inherent variability of language and the complexities and uniqueness of human speech. We see the promising potential in recent transdisciplinary work that incorporates linguistic knowledge into AI approaches to provide pathways for expert-in-the-loop and to move beyond expert agnostic AI-based methods for more robust and comprehensive deepfake detection.

Early identification of respiratory irregularities is critical for improving lung health and reducing global mortality rates. The analysis of respiratory sounds plays a significant role in characterizing the respiratory system's condition and identifying abnormalities. The main contribution of this study is to investigate the performance when the input data, represented by cochleogram, is used to feed the Vision Transformer architecture, since this input classifier combination is the first time it has been applied to adventitious sound classification to our knowledge. Although ViT has shown promising results in audio classification tasks by applying self attention to spectrogram patches, we extend this approach by applying the cochleogram, which captures specific spectro-temporal features of adventitious sounds. The proposed methodology is evaluated on the ICBHI dataset. We compare the classification performance of ViT with other state of the art CNN approaches using spectrogram, Mel frequency cepstral coefficients, constant Q transform, and cochleogram as input data. Our results confirm the superior classification performance combining cochleogram and ViT, highlighting the potential of ViT for reliable respiratory sound classification. This study contributes to the ongoing efforts in developing automatic intelligent techniques with the aim to significantly augment the speed and effectiveness of respiratory disease detection, thereby addressing a critical need in the medical field.