论文检索

Jixun Yao, Yang Yuguang, Yu Pan, Ziqian Ning, Jianhao Ye, Hongbin Zhou, Lei Xie

Zero-shot voice conversion (VC) aims to transfer the timbre from the source speaker to an arbitrary unseen speaker while preserving the original linguistic content. Despite recent advancements in zero-shot VC using language model-based or diffusion-based approaches, several challenges remain: 1) current approaches primarily focus on adapting timbre from unseen speakers and are unable to transfer style and timbre to different unseen speakers independently; 2) these approaches often suffer from slower inference speeds due to the autoregressive modeling methods or the need for numerous sampling steps; 3) the quality and similarity of the converted samples are still not fully satisfactory. To address these challenges, we propose a Style controllable zero-shot VC approach named StableVC, which aims to transfer timbre and style from source speech to different unseen target speakers. Specifically, we decompose speech into linguistic content, timbre, and style, and then employ a conditional flow matching module to reconstruct the high-quality mel-spectrogram based on these decomposed features. To effectively capture timbre and style in a zero-shot manner, we introduce a novel dual attention mechanism with an adaptive gate, rather than using conventional feature concatenation. With this non-autoregressive design, StableVC can efficiently capture the intricate timbre and style from different unseen speakers and generate high-quality speech significantly faster than real-time. Experiments demonstrate that our proposed StableVC outperforms state-of-the-art baseline systems in zero-shot VC and achieves flexible control over timbre and style from different unseen speakers. Moreover, StableVC offers approximately 25x and 1.65x faster sampling compared to autoregressive and diffusion-based baselines.

Yihan Wu, Yichen Lu, Yifan Peng, Xihua Wang, Ruihua Song, Shinji Watanabe

Audiovisual Automatic Speech Recognition (AV-ASR) aims to improve speech recognition accuracy by leveraging visual signals. It is particularly challenging in unconstrained real-world scenarios across various domains due to noisy acoustic environments, spontaneous speech, and the uncertain use of visual information. Most previous works fine-tune audio-only ASR models on audiovisual datasets, optimizing them for conventional ASR objectives. However, they often neglect visual features and common errors in unconstrained video scenarios. In this paper, we propose using a preference optimization strategy to improve speech recognition accuracy for real-world videos. First, we create preference data via simulating common errors that occurred in AV-ASR from two focals: manipulating the audio or vision input and rewriting the output transcript. Second, we propose BPO-AVASR, a Bifocal Preference Optimization method to improve AV-ASR models by leveraging both input-side and output-side preference. Extensive experiments demonstrate that our approach significantly improves speech recognition accuracy across various domains, outperforming previous state-of-the-art models on real-world video speech recognition.

Linqin Wang, Yaping Liu, Zhengtao Yu, Shengxiang Gao, Cunli Mao, Yuxin Huang, Wenjun Wang, Ling Dong

With the rapid advancement of large language models (LLMs), discrete speech representations have become crucial for integrating speech into LLMs. Existing methods for speech representation discretization rely on a predefined codebook size and Euclidean distance-based quantization. However, 1) the size of codebook is a critical parameter that affects both codec performance and downstream task training efficiency. 2) The Euclidean distance-based quantization may lead to audio distortion when the size of the codebook is controlled within a reasonable range. In fact, in the field of information compression, structural information and entropy guidance are crucial, but previous methods have largely overlooked these factors. Therefore, we address the above issues from an information-theoretic perspective, we present SECodec, a novel speech representation codec based on structural entropy (SE) for building speech language models. Specifically, we first model speech as a graph, clustering the speech features nodes within the graph and extracting the corresponding codebook by hierarchically and disentangledly minimizing 2D SE. Then, to address the issue of audio distortion, we propose a new quantization method. This method still adheres to the 2D SE minimization principle, adaptively selecting the most suitable token corresponding to the cluster for each incoming original speech node. Furthermore, we develop a Structural Entropy-based Speech Language Model (SESLM) that leverages SECodec. Experimental results demonstrate that SECodec performs comparably to EnCodec in speech reconstruction, and SESLM surpasses VALL-E in zero-shot text-to-speech tasks.

Yakun Song, Zhuo Chen, Xiaofei Wang, Ziyang Ma, Xie Chen

The language model (LM) approach based on acoustic and linguistic prompts, such as VALL-E, has achieved remarkable progress in the field of zero-shot audio generation. However, existing methods still have some limitations: 1) repetitions, transpositions, and omissions in the output synthesized speech due to limited alignment constraints between audio and phoneme tokens; 2) challenges of fine-grained control over the synthesized speech with autoregressive (AR) language model; 3) infinite silence generation due to the nature of AR-based decoding, especially under the greedy strategy. To alleviate these issues, we propose ELLA-V, a simple but efficient LM-based zero-shot text-to-speech (TTS) framework, which enables fine-grained control over synthesized audio at the phoneme level. The key to ELLA-V is interleaving sequences of acoustic and phoneme tokens, where phoneme tokens appear ahead of the corresponding acoustic tokens. The experimental findings reveal that our model outperforms baselines in terms of accuracy and delivers more stable results using both greedy and sampling-based decoding strategies.

Yan Rong, Li Liu

Face-based Voice Conversion (FVC) is a novel task that leverages facial images to generate the target speaker's voice style. Previous work has two shortcomings: (1) suffering from obtaining facial embeddings that are well-aligned with the speaker's voice identity information, and (2) inadequacy in decoupling content and speaker identity information from the audio input. To address these issues, we present a novel FVC method, Identity-Disentanglement Face-based Voice Conversion (ID-FaceVC), which overcomes the above two limitations. More precisely, we propose an Identity-Aware Query-based Contrastive Learning (IAQ-CL) module to extract speaker-specific facial features, and a Mutual Information-based Dual Decoupling (MIDD) module to purify content features from audio, ensuring clear and high-quality voice conversion. Besides, unlike prior works, our method can accept either audio or text inputs, offering controllable speech generation with adjustable emotional tone and speed. Extensive experiments demonstrate that ID-FaceVC achieves state-of-the-art performance across various metrics, with qualitative and user study results confirming its effectiveness in naturalness, similarity, and diversity.

Tingyu Mo, Jacqueline C. K. Lam, Victor O. K. Li, Lawrence Y. L. Cheung

Alzheimer’s Disease (AD) is an irreversible neurodegenerative disease affecting 50 million people worldwide. Low-cost, accurate identification of key markers of AD is crucial for timely diagnosis and intervention. Language impairment is one of the earliest signs of cognitive decline, which can be used to discriminate AD patients from normal control (NC) individuals. Patient-interviewer dialogues may be used to detect such impairments, but they are often mixed with ambiguous, noisy, and irrelevant information, making the AD detection task difficult. Moreover, the limited availability of AD speech samples and variability in their speech styles pose significant challenges in developing robust speech-based AD detection models. To address these challenges, we propose DECT, a novel speech-based domain-specific approach leveraging large language models (LLMs) for fine-grained linguistic analysis and label-switched label-preserved (LSLP) data generation. Our study presents four novelties: (1) We harness the summarizing capabilities of LLMs to identify and distill key Cognitive-Linguistic (CL) information (atoms) from noisy speech transcripts, effectively filtering irrelevant information. (2) We leverage the inherent linguistic knowledge of LLMs to extract linguistic markers from unstructured and heterogeneous audio transcripts. (3) We exploit the compositional ability of LLMs to generate LSLP AD speech transcripts consisting of diverse linguistic patterns to overcome the speech data scarcity challenge and enhance the robustness of AD detection models. (4) We use the augmented AD textual speech transcript dataset and a more fine-grained representation of AD textual speech transcript data to fine-tune the AD detection model. The results have shown that DECT, an integrated, LLM-assisted, speech-based AD detection model demonstrates superior model performance with an 11% improvement in AD detection accuracy on the datasets from DementiaBank compared to the baselines.

Hao Ming, Xinguo Yu, Xiaotian Cheng, Zhenquan Shen, Xiaopan Lyu

Large language models (LLMs) have made significant advancements in math problem solving, but their large size and high latency render them impractical for real-world applications in intelligent mathematics solvers. Recently, task-agnostic compact models have been developed to replace LLMs in general natural language processing tasks. However, these models often struggle to acquire sufficient math-related knowledge from LLMs, leading to unsatisfactory performance in solving math word problems (MWPs). To develop a specialized compact model for representing MWPs, we develop the knowledge distillation (KD) technique to extract mathematical semantics knowledge from the large pre-trained model BERT. Effective knowledge types and distillation strategies are explored through extensive experiments. Our KD algorithm employs multi-knowledge distillation to extract fundamental knowledge from hidden states in the middle to lower layers, while also incorporating knowledge of mathematical relations and symbol constraints from higher-layer outputs and math decoder outputs, by leveraging bottleneck networks. Pre-training tasks on MWP datasets, such as masked language modeling and part-of-speech tagging, are also utilized to enhance the generalization of the compact model for MWP understanding. Additionally, a simple parameter mixing strategy is employed to prevent catastrophic forgetting of acquired knowledge. Our findings indicate that our approach can reduce the size of a BERT model by 10% while retaining approximately 95% of its performance on MWP datasets, outperforming the mainstream BERT-based task-agnostic compact models. The efficacy of each component has been validated through ablation studies.

Ziyang Ma, Guanrou Yang, Yifan Yang, Zhifu Gao, Jiaming Wang, Zhihao Du, Fan Yu, Qian Chen, Siqi Zheng, Shiliang Zhang 等

In this paper, we focus on prompting one of the most important tasks in the field of speech processing, i.e., automatic speech recognition (ASR), with speech foundation encoders and large language models (LLM). Despite the growing body of research in this area, we find that many crucial design decisions in LLM-based ASR systems are often inadequately justified. This lack of clarity impedes the field's progress, making it challenging to pinpoint which design choices truly improve model performance. To address these challenges, we conduct a comprehensive series of experiments that explore various aspects, leading to the optimal LLM-based ASR system. We found that delicate designs are not necessary, while a clean setup with little task-specific design is competent. The models achieve strong performance on the Librispeech and Gigaspeech datasets, compared to both LLM-based models and non-LLM-based models. Finally, we explore the capability emergence of LLM-based ASR in the process of modal alignment. We hope that our study can facilitate the research on extending LLM with cross-modality capacity and shed light on the LLM-based ASR community.

Ziyang Ma, Yakun Song, Chenpeng Du, Jian Cong, Zhuo Chen, Yuping Wang, Yuxuan Wang, Xie Chen

Dialogue serves as the most natural manner of human-computer interaction (HCI). Recent advancements in speech language models (SLM), have significantly enhanced speech-based conversational AI. However, these models are limited to turn-based conversation, lacking the ability to interact with humans in real-time spoken scenarios, for example, being interrupted when the generated content is not satisfactory. To address these limitations, we explore full duplex modeling (FDM) in interactive speech language models (iSLM), focusing on enhancing real-time interaction and, more explicitly, exploring the quintessential ability of interruption. We introduce a novel model design, namely listening-while-speaking language model (LSLM), an end-to-end system equipped with both listening and speaking channels. Our LSLM employs a token-based decoder-only TTS for speech generation and a streaming self-supervised learning (SSL) encoder for real-time audio input. LSLM fuses both channels for autoregressive generation and detects turn-taking in real time. Three fusion strategies—early fusion, middle fusion, and late fusion—are explored, with middle fusion achieving an optimal balance between speech generation and real-time interaction. Two experimental settings, command-based FDM and voice-based FDM, demonstrate LSLM’s robustness to noise and sensitivity to diverse instructions. Our results highlight LSLM’s capability to achieve duplex communication with minimal impact on existing systems. This study aims to advance the development of interactive speech dialogue systems, enhancing their applicability in real-world contexts.

Rui Liu, Shuwei He, Yifan Hu, Haizhou Li

Visual Text-to-Speech (VTTS) aims to take the environmental image as the prompt to synthesize the reverberant speech for the spoken content. The challenge of this task lies in understanding the spatial environment from the image. Many attempts have been made to extract global spatial visual information from the RGB space of an spatial image. However, local and depth image information are crucial for understanding the spatial environment, which previous works have ignored. To address the issues, we propose a novel multi-modal and multi-scale spatial environment understanding scheme to achieve immersive VTTS, termed M2SE-VTTS. The multi-modal aims to take both the RGB and Depth spaces of the spatial image to learn more comprehensive spatial information, and the multi-scale seeks to model the local and global spatial knowledge simultaneously. Specifically, we first split the RGB and Depth images into patches and adopt the Gemini-generated environment captions to guide the local spatial understanding. After that, the multi-modal and multi-scale features are integrated by the local-aware global spatial understanding. In this way, M2SE-VTTS effectively models the interactions between local and global spatial contexts in the multi-modal spatial environment. Objective and subjective evaluations suggest that our model outperforms the advanced baselines in environmental speech generation.

Jiaqing Liu, Chong Deng, Qinglin Zhang, Shilin Zhou, Qian Chen, Hai Yu, Wen Wang

Automatic Speech Recognition (ASR) transcripts exhibit recognition errors and various spoken language phenomena such as disfluencies, ungrammatical sentences, and incomplete sentences, hence suffering from poor readability. To improve readability, we propose a Contextualized Spoken-to-Written conversion (CoS2W) task to address ASR and grammar errors and also transfer the informal text into the formal style with content preserved, utilizing contexts and auxiliary information. This task naturally matches the in-context learning capabilities of Large Language Models (LLMs). To facilitate comprehensive comparisons of various LLMs, we construct a document-level Spoken-to-Written conversion of ASR Transcripts Benchmark (SWAB) dataset. Using SWAB, we study the impact of different granularity levels on the CoS2W performance, and propose methods to exploit contexts and auxiliary information to enhance the outputs. Experimental results reveal that LLMs have the potential to excel in the CoS2W task, particularly in grammaticality and formality, our methods achieve effective understanding of contexts and auxiliary information by LLMs. We further investigate the effectiveness of using LLMs as evaluators and find that LLM evaluators show strong correlations with human evaluations on rankings of faithfulness and formality, which validates the reliability of LLM evaluators for the CoS2W task.

Qi Lian, Yu Qi, Yueming Wang

Denoising diffusion probabilistic models (DDPMs) have gained popularity in devising neural vocoders and obtained outstanding performance. However, existing DDPM-based neural vocoders struggle to handle the prosody diversities due to their susceptibility to mode-collapse issues confronted with imbalanced data. We introduced Cauchy Diffusion, a model incorporating the Cauchy noises to address this challenge. The heavy-tailed Cauchy distribution exhibits better resilience to imbalanced speech data, potentially improving prosody modeling. Our experiments on the LJSpeech and VCTK datasets demonstrate that Cauchy Diffusion achieved state-of-the-art speech synthesis performance. Compared to existing neural vocoders, our Cauchy Diffusion notably improved speech diversity while maintaining superior speech quality. Remarkably, Cauchy Diffusion surpassed neural vocoders based on generative adversarial networks (GANs) that are explicitly optimized to improve diversity.

Linfei Li, Lin Zhang, Zhong Wang, Fengyi Zhang, Zelin Li, Ying Shen

Although Coordinate-MLP-based implicit neural representations have excelled in representing radiance fields, 3D shapes, and images, their application to audio signals remains underexplored. To fill this gap, we investigate existing implicit neural representations, from which we extract 3 types of positional encoding and 16 commonly used activation functions. Through combinatorial design, we establish the first benchmark for Coordinate-MLPs in audio signal representations. Our benchmark reveals that Coordinate-MLPs require complex hyperparameter tuning and frequency-dependent initialization, limiting their robustness. To address these issues, we propose Fourier-ASR, a novel framework based on the Fourier series theorem and the Kolmogorov-Arnold representation theorem. Fourier-ASR introduces Fourier Kolmogorov-Arnold Networks (Fourier-KAN), which leverage periodicity and strong nonlinearity to represent audio signals, eliminating the need for additional positional encoding. Furthermore, a Frequency-adaptive Learning Strategy (FaLS) is proposed to enhance the convergence of Fourier-KAN by capturing high-frequency components and preventing overfitting of low-frequency signals. Extensive experiments conducted on natural speech and music datasets reveal that: (1) well-designed positional encoding and activation functions in Coordinate-MLPs can effectively improve audio representation quality; and (2) Fourier-ASR can robustly represent complex audio signals without extensive hyperparameter tuning. Looking ahead, the continuity and infinite resolution of implicit audio representations make our research highly promising for tasks such as audio compression, synthesis, and generation.

Sangmin Lee, Woojin Chung, Hong-Goo Kang

Building a universal multilingual automatic speech recognition (ASR) model that performs equitably across languages has long been a challenge due to its inherent difficulties. To address this task we introduce a Language-Agnostic Multilingual ASR pipeline through orthography Unification and language-specific Transliteration (LAMA-UT). LAMA-UT operates without any language-specific modules while matching the performance of state-of-the-art models trained on a minimal amount of data. Our pipeline consists of two key steps. First, we utilize a universal transcription generator to unify orthographic features into Romanized form and capture common phonetic characteristics across diverse languages. Second, we utilize a universal converter to transform these universal transcriptions into language-specific ones. In experiments, we demonstrate the effectiveness of our proposed method leveraging universal transcriptions for massively multilingual ASR. Our pipeline achieves a relative error reduction rate of 45% when compared to Whisper and performs comparably to MMS, despite being trained on only 0.1% of Whisper's training data. Furthermore, our pipeline does not rely on any language-specific modules. However, it performs on par with zero-shot ASR approaches which utilize additional language-specific lexicons and language models. We expect this framework to serve as a cornerstone for flexible multilingual ASR systems that are generalizable even to unseen languages.

Eungbeom Kim, Kyogu Lee

Uncertainty estimation has been widely applied for trustworthy automatic speech recognition (ASR) systems across training and inference stages. In the training stage, previous studies show that uncertainty can facilitate self-training by filtering out unlabeled data samples with high uncertainty. However, the current sequence-level uncertainty estimation method for connectionist temporal classification (CTC) based ASR models drops the output probability information and depends only on the textual distance of decoded predictions. In this study, we argue that this results in limited performance improvement and propose a novel output probability-based sequence-level uncertainty estimation method. We also categorize uncertainty as pseudo-label uncertainty and in-training uncertainty for the self-training process. Finally, we present uncertainty-aware self-training for CTC-based ASR models and experimentally show the effectiveness of the proposed method compared to the baselines.

Shengpeng Ji, Ziyue Jiang, Jialong Zuo, Minghui Fang, Yifu Chen, Tao Jin, Zhou Zhao

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

Sieun Hyeon, Kyudan Jung, Jaehee Won, Nam-Joon Kim, Hyun Gon Ryu, Hyuk-Jae Lee, Jaeyoung Do

In various academic and professional settings, such as mathematics lectures or research presentations, it is often necessary to convey mathematical expressions orally. However, reading mathematical expressions aloud without accompanying visuals can significantly hinder comprehension, especially for those who are hearing-impaired or rely on subtitles due to language barriers. For instance, when a presenter reads Euler's Formula, current Automatic Speech Recognition (ASR) models often produce a verbose and error-prone textual description (e.g., e to the power of i x equals cosine of x plus i 'side' of x), instead of the concise LaTeX format, which hampers clear understanding and communication. To address this issue, we introduce MathSpeech, a novel pipeline that integrates ASR models with small Language Models (sLMs) to correct errors in mathematical expressions and accurately convert spoken expressions into structured LaTeX representations. Evaluated on a new dataset derived from lecture recordings, MathSpeech demonstrates LaTeX generation capabilities comparable to leading commercial Large Language Models (LLMs), while leveraging fine-tuned small language models of only 120M parameters. Specifically, in terms of CER, BLEU, and ROUGE scores for LaTeX translation, MathSpeech demonstrated significantly superior capabilities compared to GPT-4o. We observed a decrease in CER from 0.390 to 0.298, and higher ROUGE/BLEU scores compared to GPT-4o.

Xiangheng He, Junjie Chen, Zixing Zhang, Björn Schuller

Prosody contains rich information beyond the literal meaning of words, which is crucial for the intelligibility of speech. Current models still fall short in phrasing and intonation; they not only miss or misplace breaks when synthesizing long sentences with complex structures but also produce unnatural intonation. We propose ProsodyFM, a prosody-aware text-to-speech synthesis (TTS) model with a flow-matching (FM) backbone that aims to enhance the phrasing and intonation aspects of prosody. ProsodyFM introduces two key components: a Phrase Break Encoder to capture initial phrase break locations, followed by a Duration Predictor for the flexible adjustment of break durations; and a Terminal Intonation Encoder which learns a bank of intonation shape tokens combined with a novel Pitch Processor for more robust modeling of human-perceived intonation change. ProsodyFM is trained with no explicit prosodic labels and yet can uncover a broad spectrum of break durations and intonation patterns. Experimental results demonstrate that ProsodyFM can effectively improve the phrasing and intonation aspects of prosody, thereby enhancing the overall intelligibility compared to four state-of-the-art (SOTA) models. Out-of-distribution experiments show that this prosody improvement can further bring ProsodyFM superior generalizability for unseen complex sentences and speakers. Our case study intuitively illustrates the powerful and fine-grained controllability of ProsodyFM over phrasing and intonation.

Shoutao Guo, Shaolei Zhang, Zhengrui Ma, Yang Feng

Simultaneous generation models write generation results while reading streaming inputs, necessitating a policy-maker to determine the appropriate output timing. Existing simultaneous generation methods generally adopt the traditional encoder-decoder architecture and learn the generation and policy-making capabilities through complex dynamic programming techniques. Although LLMs excel at text generation, they face challenges in taking on the role of policy-makers through traditional training methods, limiting their exploration in simultaneous generation. To overcome these limitations, we propose a novel LLM-driven Simultaneous Generation (LSG) framework, which allows the off-the-shelf LLM to decide the generation timing and produce output concurrently. Specifically, LSG selects the generation policy that minimizes latency as the baseline policy. Referring to the baseline policy, LSG enables the LLM to devise an improved generation policy that better balances latency and generation quality, and writes generation results accordingly. Experiments on simultaneous translation and streaming automatic speech recognition tasks show that our method can achieve state-of-the-art performance utilizing the open-source LLMs and demonstrate practicality in real-world scenarios.

Ashishkumar Prabhakar Gudmalwar, Ishan Darshan Biyani, Nirmesh J. Shah, Pankaj Wasnik, Rajiv Ratn Shah

The Emotional Voice Conversion (EVC) aims to convert the discrete emotional state from the source emotion to the target for a given speech utterance while preserving linguistic content. In this paper, we propose regularizing emotion intensity in the diffusion-based EVC framework to generate precise speech of the target emotion. Traditional approaches control the intensity of an emotional state in the utterance via emotion class probabilities or intensity labels that often lead to inept style manipulations and degradations in quality. On the contrary, we aim to regulate emotion intensity using self-supervised learning-based feature representations and unsupervised directional latent vector modeling (DVM) in the emotional embedding space within a diffusion-based framework. These emotion embeddings can be modified based on the given target emotion intensity and the corresponding direction vector. Furthermore, the updated embeddings can be fused in the reverse diffusion process to generate the speech with the desired emotion and intensity. In summary, this paper aims to achieve high-quality emotional intensity regularization in the diffusion-based EVC framework, which is the first of its kind work. The effectiveness of the proposed method has been shown across state-of-the-art (SOTA) baselines in terms of subjective and objective evaluations for the English and Hindi languages.