Xie Chen

Yexing Du, Youcheng Pan, Ziyang Ma et al.

Multimodal Large Language Models (MLLMs) have achieved significant success in Speech-to-Text Translation (S2TT) tasks. While most existing research has focused on English-centric translation directions, the exploration of many-to-many translation is still limited by the scarcity of parallel data. To address this, we propose a three-stage curriculum learning strategy that leverages the machine translation capabilities of large language models and adapts them to S2TT tasks, enabling effective learning in low-resource settings. We trained MLLMs with varying parameter sizes (3B, 7B, and 32B) and evaluated the proposed strategy using the FLEURS and CoVoST-2 datasets. Experimental results show that the proposed strategy achieves state-of-the-art average performance in $15\times14$ language pairs, requiring fewer than 10 hours of speech data per language to achieve competitive results. The source code and models are released at https://github.com/yxduir/LLM-SRT.

Mingyu Cui, Yifan Yang, Jiajun Deng et al.

Self-supervised learning (SSL) based discrete speech representations are highly compact and domain adaptable. In this paper, SSL discrete speech features extracted from WavLM models are used as additional cross-utterance acoustic context features in Zipformer-Transducer ASR systems. The efficacy of replacing Fbank features with discrete token features for modelling either cross-utterance contexts (from preceding and future segments), or current utterance's internal contexts alone, or both at the same time, are demonstrated thoroughly on the Gigaspeech 1000-hr corpus. The best Zipformer-Transducer system using discrete tokens based cross-utterance context features outperforms the baseline using utterance internal context only with statistically significant word error rate (WER) reductions of 0.32% to 0.41% absolute (2.78% to 3.54% relative) on the dev and test data. The lowest published WER of 11.15% and 11.14% were obtained on the dev and test sets. Our work is open-source and publicly available at https://github.com/open-creator/icefall/tree/master/egs/gigaspeech/Context\_ASR.

Guanrou Yang, Ziyang Ma, Zhisheng Zheng et al.

Recent years have witnessed significant advancements in self-supervised learning (SSL) methods for speech-processing tasks. Various speech-based SSL models have been developed and present promising performance on a range of downstream tasks including speech recognition. However, existing speech-based SSL models face a common dilemma in terms of computational cost, which might hinder their potential application and in-depth academic research. To address this issue, we first analyze the computational cost of different modules during HuBERT pre-training and then introduce a stack of efficiency optimizations, which is named Fast-HuBERT in this paper. The proposed Fast-HuBERT can be trained in 1.1 days with 8 V100 GPUs on the Librispeech 960h benchmark, without performance degradation, resulting in a 5.2x speedup, compared to the original implementation. Moreover, we explore two well-studied techniques in the Fast-HuBERT and demonstrate consistent improvements as reported in previous work.

Chen Xie, Ciyun Lin, Xiaoyu Zheng et al.

Understanding and predicting pedestrian crossing behavioral intention is crucial for the driving safety of autonomous vehicles. Nonetheless, challenges emerge when using promising images or environmental context masks to extract various factors for time-series network modeling, causing pre-processing errors or a loss of efficiency. Typically, pedestrian positions captured by onboard cameras are often distorted and do not accurately reflect their actual movements. To address these issues, GTransPDM -- a Graph-embedded Transformer with a Position Decoupling Module -- was developed for pedestrian crossing intention prediction by leveraging multi-modal features. First, a positional decoupling module was proposed to decompose pedestrian lateral motion and encode depth cues in the image view. Then, a graph-embedded Transformer was designed to capture the spatio-temporal dynamics of human pose skeletons, integrating essential factors such as position, skeleton, and ego-vehicle motion. Experimental results indicate that the proposed method achieves 92% accuracy on the PIE dataset and 87% accuracy on the JAAD dataset, with a processing speed of 0.05ms. It outperforms the state-of-the-art in comparison.

Xinsheng Wang, Mingqi Jiang, Ziyang Ma et al.

Recent advancements in large language models (LLMs) have driven significant progress in zero-shot text-to-speech (TTS) synthesis. However, existing foundation models rely on multi-stage processing or complex architectures for predicting multiple codebooks, limiting efficiency and integration flexibility. To overcome these challenges, we introduce Spark-TTS, a novel system powered by BiCodec, a single-stream speech codec that decomposes speech into two complementary token types: low-bitrate semantic tokens for linguistic content and fixed-length global tokens for speaker attributes. This disentangled representation, combined with the Qwen2.5 LLM and a chain-of-thought (CoT) generation approach, enables both coarse-grained control (e.g., gender, speaking style) and fine-grained adjustments (e.g., precise pitch values, speaking rate). To facilitate research in controllable TTS, we introduce VoxBox, a meticulously curated 100,000-hour dataset with comprehensive attribute annotations. Extensive experiments demonstrate that Spark-TTS not only achieves state-of-the-art zero-shot voice cloning but also generates highly customizable voices that surpass the limitations of reference-based synthesis. Source code, pre-trained models, and audio samples are available at https://github.com/SparkAudio/Spark-TTS.

Yiwei Guo, Zhihan Li, Junjie Li et al.

We propose a new speech discrete token vocoder, vec2wav 2.0, which advances voice conversion (VC). We use discrete tokens from speech self-supervised models as the content features of source speech, and treat VC as a prompted vocoding task. To amend the loss of speaker timbre in the content tokens, vec2wav 2.0 utilizes the WavLM features to provide strong timbre-dependent information. A novel adaptive Snake activation function is proposed to better incorporate timbre into the waveform reconstruction process. In this way, vec2wav 2.0 learns to alter the speaker timbre appropriately given different reference prompts. Also, no supervised data is required for vec2wav 2.0 to be effectively trained. Experimental results demonstrate that vec2wav 2.0 outperforms all other baselines to a considerable margin in terms of audio quality and speaker similarity in any-to-any VC. Ablation studies verify the effects made by the proposed techniques. Moreover, vec2wav 2.0 achieves competitive cross-lingual VC even only trained on monolingual corpus. Thus, vec2wav 2.0 shows timbre can potentially be manipulated only by speech token vocoders, pushing the frontiers of VC and speech synthesis.

Bohan Li, Feiyu Shen, Yiwei Guo et al.

Discretizing speech into tokens and generating them by a decoder-only model have been a promising direction for text-to-speech (TTS) and spoken language modeling (SLM). To shorten the sequence length of speech tokens, acoustic byte-pair encoding (BPE) has emerged in SLM that treats speech tokens from self-supervised semantic representations as characters to further compress the token sequence. But the gain in TTS has not been fully investigated, and the proper choice of acoustic BPE remains unclear. In this work, we conduct a comprehensive study on various settings of acoustic BPE to explore its effectiveness in decoder-only TTS models with semantic speech tokens. Experiments on LibriTTS verify that acoustic BPE uniformly increases the intelligibility and diversity of synthesized speech, while showing different features across BPE settings. Hence, acoustic BPE is a favorable tool for decoder-only TTS.

Mingyu Cui, Daxin Tan, Yifan Yang et al.

With the advancement of Self-supervised Learning (SSL) in speech-related tasks, there has been growing interest in utilizing discrete tokens generated by SSL for automatic speech recognition (ASR), as they offer faster processing techniques. However, previous studies primarily focused on multilingual ASR with Fbank features or English ASR with discrete tokens, leaving a gap in adapting discrete tokens for multilingual ASR scenarios. This study presents a comprehensive comparison of discrete tokens generated by various leading SSL models across multiple language domains. We aim to explore the performance and efficiency of speech discrete tokens across multiple language domains for both monolingual and multilingual ASR scenarios. Experimental results demonstrate that discrete tokens achieve comparable results against systems trained on Fbank features in ASR tasks across seven language domains with an average word error rate (WER) reduction of 0.31% and 1.76% absolute (2.80% and 15.70% relative) on dev and test sets respectively, with particularly WER reduction of 6.82% absolute (41.48% relative) on the Polish test set.

Guanrou Yang, Chen Yang, Qian Chen et al.

Human speech goes beyond the mere transfer of information; it is a profound exchange of emotions and a connection between individuals. While Text-to-Speech (TTS) models have made huge progress, they still face challenges in controlling the emotional expression in the generated speech. In this work, we propose EmoVoice, a novel emotion-controllable TTS model that exploits large language models (LLMs) to enable fine-grained freestyle natural language emotion control, and a phoneme boost variant design that makes the model output phoneme tokens and audio tokens in parallel to enhance content consistency, inspired by chain-of-thought (CoT) and chain-of-modality (CoM) techniques. Besides, we introduce EmoVoice-DB, a high-quality 40-hour English emotion dataset featuring expressive speech and fine-grained emotion labels with natural language descriptions. EmoVoice achieves state-of-the-art performance on the English EmoVoice-DB test set using only synthetic training data, and on the Chinese Secap test set using our in-house data. We further investigate the reliability of existing emotion evaluation metrics and their alignment with human perceptual preferences, and explore using SOTA multimodal LLMs GPT-4o-audio and Gemini to assess emotional speech. Dataset, code, checkpoints, and demo samples are available at https://github.com/yanghaha0908/EmoVoice.

Ruiqi Yan, Xiquan Li, Wenxi Chen et al.

Recent advances in large language models (LLMs) have driven significant progress in end-to-end spoken dialogue models (SDMs). In contrast to text-based LLMs, the evaluation framework for SDMs should encompass both cognitive dimensions (e.g., logical reasoning, knowledge) and speech-related aspects (e.g., paralinguistic cues, audio quality). However, there is still a lack of comprehensive evaluations for SDMs in speech-to-speech (S2S) scenarios. To address this gap, we propose URO-Bench, an extensive benchmark for SDMs. Notably, URO-Bench is the first S2S benchmark that covers evaluations about multilingualism, multi-round dialogues, and paralinguistics. Our benchmark is divided into two difficulty levels: basic track and pro track, each comprising 20 test sets, evaluating the spoken dialogue model's abilities in Understanding, Reasoning, and Oral conversation. Evaluations on our proposed benchmark reveal that current open-source SDMs perform rather well in daily QA tasks, but lag behind their backbone LLMs in terms of instruction-following ability and also suffer from catastrophic forgetting. Their performance in advanced evaluations of paralinguistic information and audio understanding remains subpar, highlighting the need for further research in this direction. We hope that URO-Bench can facilitate the development of spoken dialogue models by providing a multifaceted evaluation of existing models and helping to track progress in this area.

Yakun Song, Jiawei Chen, Xiaobin Zhuang et al.

Neural audio codecs have made significant strides in efficiently mapping raw audio waveforms into discrete token representations, which are foundational for contemporary audio generative models. However, most existing codecs are optimized primarily for reconstruction quality, often at the expense of the downstream modelability of the encoded tokens. Motivated by the need to overcome this bottleneck, we introduce $\textbf{MagiCodec}$, a novel single-layer, streaming Transformer-based audio codec. MagiCodec is designed with a multistage training pipeline that incorporates Gaussian noise injection and latent regularization, explicitly targeting the enhancement of semantic expressiveness in the generated codes while preserving high reconstruction fidelity. We analytically derive the effect of noise injection in the frequency domain, demonstrating its efficacy in attenuating high-frequency components and fostering robust tokenization. Extensive experimental evaluations show that MagiCodec surpasses state-of-the-art codecs in both reconstruction quality and downstream tasks. Notably, the tokens produced by MagiCodec exhibit Zipf-like distributions, as observed in natural languages, thereby improving compatibility with language-model-based generative architectures. The code and pre-trained models are available at https://github.com/Ereboas/MagiCodec.

Yakun Song, Zhuo Chen, Xiaofei Wang et al.

Neural codec language model (LM) has demonstrated strong capability in zero-shot text-to-speech (TTS) synthesis. However, the codec LM often suffers from limitations in inference speed and stability, due to its auto-regressive nature and implicit alignment between text and audio. In this work, to handle these challenges, we introduce a new variant of neural codec LM, namely TacoLM. Specifically, TacoLM introduces a gated attention mechanism to improve the training and inference efficiency and reduce the model size. Meanwhile, an additional gated cross-attention layer is included for each decoder layer, which improves the efficiency and content accuracy of the synthesized speech. In the evaluation of the Librispeech corpus, the proposed TacoLM achieves a better word error rate, speaker similarity, and mean opinion score, with 90% fewer parameters and 5.2 times speed up, compared with VALL-E. Demo and code is available at https://ereboas.github.io/TacoLM/.

Guanrou Yang, Ziyang Ma, Zhifu Gao et al.

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.

Xiquan Li, Wenxi Chen, Ziyang Ma et al.

While automated audio captioning (AAC) has made notable progress, traditional fully supervised AAC models still face two critical challenges: the need for expensive audio-text pair data for training and performance degradation when transferring across domains. To overcome these limitations, we present DRCap, a data-efficient and flexible zero-shot audio captioning system that requires text-only data for training and can quickly adapt to new domains without additional fine-tuning. DRCap integrates a contrastive language-audio pre-training (CLAP) model and a large-language model (LLM) as its backbone. During training, the model predicts the ground-truth caption with a fixed text encoder from CLAP, whereas, during inference, the text encoder is replaced with the audio encoder to generate captions for audio clips in a zero-shot manner. To mitigate the modality gap of the CLAP model, we use both the projection strategy from the encoder side and the retrieval-augmented generation strategy from the decoder side. Specifically, audio embeddings are first projected onto a text embedding support to absorb extensive semantic information within the joint multi-modal space of CLAP. At the same time, similar captions retrieved from a datastore are fed as prompts to instruct the LLM, incorporating external knowledge to take full advantage of its strong generative capability. Conditioned on both the projected CLAP embedding and the retrieved similar captions, the model is able to produce a more accurate and semantically rich textual description. By tailoring the text embedding support and the caption datastore to the target domain, DRCap acquires a robust ability to adapt to new domains in a training-free manner. Experimental results demonstrate that DRCap outperforms all other zero-shot models in in-domain scenarios and achieves state-of-the-art performance in cross-domain scenarios.

Tao Liu, Ziyang Ma, Qi Chen et al.

We present VQTalker, a Vector Quantization-based framework for multilingual talking head generation that addresses the challenges of lip synchronization and natural motion across diverse languages. Our approach is grounded in the phonetic principle that human speech comprises a finite set of distinct sound units (phonemes) and corresponding visual articulations (visemes), which often share commonalities across languages. We introduce a facial motion tokenizer based on Group Residual Finite Scalar Quantization (GRFSQ), which creates a discretized representation of facial features. This method enables comprehensive capture of facial movements while improving generalization to multiple languages, even with limited training data. Building on this quantized representation, we implement a coarse-to-fine motion generation process that progressively refines facial animations. Extensive experiments demonstrate that VQTalker achieves state-of-the-art performance in both video-driven and speech-driven scenarios, particularly in multilingual settings. Notably, our method achieves high-quality results at a resolution of 512*512 pixels while maintaining a lower bitrate of approximately 11 kbps. Our work opens new possibilities for cross-lingual talking face generation. Synthetic results can be viewed at https://x-lance.github.io/VQTalker.

Sen Liu, Yiwei Guo, Xie Chen et al.

While acoustic expressiveness has long been studied in expressive text-to-speech (ETTS), the inherent expressiveness in text lacks sufficient attention, especially for ETTS of artistic works. In this paper, we introduce StoryTTS, a highly ETTS dataset that contains rich expressiveness both in acoustic and textual perspective, from the recording of a Mandarin storytelling show. A systematic and comprehensive labeling framework is proposed for textual expressiveness. We analyze and define speech-related textual expressiveness in StoryTTS to include five distinct dimensions through linguistics, rhetoric, etc. Then we employ large language models and prompt them with a few manual annotation examples for batch annotation. The resulting corpus contains 61 hours of consecutive and highly prosodic speech equipped with accurate text transcriptions and rich textual expressiveness annotations. Therefore, StoryTTS can aid future ETTS research to fully mine the abundant intrinsic textual and acoustic features. Experiments are conducted to validate that TTS models can generate speech with improved expressiveness when integrating with the annotated textual labels in StoryTTS.

Bohan Li, Zhihan Li, Haoran Wang et al.

Recently, autoregressive (AR) language models have emerged as a dominant approach in speech synthesis, offering expressive generation and scalable training. However, conventional AR speech synthesis models relying on the next-token prediction paradigm often encounter significant challenges when handling long speech sequences. These models often struggle to construct stable frame-to-frame attention, leading to increased latency and degraded synthesis quality, thereby limiting their feasibility for real-time applications. To address these limitations, we introduce a novel dynamic chunk-wise autoregressive synthesis framework, termed DCAR, designed to enhance both efficiency and intelligibility robustness in AR speech generation. DCAR introduces a chunk-to-frame attention mechanism through training with multi-token prediction, enabling dynamic chunk prediction in variable speech contexts using a lightweight module trained on-policy. DCAR dynamically adjusts the token prediction span, significantly reducing the sequence length dependency while obtaining high synthesis quality. Comprehensive empirical evaluations demonstrate that DCAR substantially outperforms traditional next-token prediction models, achieving up to 72.27% intelligibility improvement and 2.61x inference speedup simultaneously on the test set. Furthermore, we conduct comprehensive analysis to support it as a versatile foundation for next-generation speech synthesis systems.

Yifan Yang, Zheshu Song, Jianheng Zhuo et al.

The evolution of speech technology has been spurred by the rapid increase in dataset sizes. Traditional speech models generally depend on a large amount of labeled training data, which is scarce for low-resource languages. This paper presents GigaSpeech 2, a large-scale, multi-domain, multilingual speech recognition corpus. It is designed for low-resource languages and does not rely on paired speech and text data. GigaSpeech 2 comprises about 30,000 hours of automatically transcribed speech, including Thai, Indonesian, and Vietnamese, gathered from unlabeled YouTube videos. We also introduce an automated pipeline for data crawling, transcription, and label refinement. Specifically, this pipeline involves Whisper for initial transcription, MMS for forced alignment, and multi-dimensional filtering for data quality assurance. A modified Noisy Student Training is developed to further refine flawed pseudo labels iteratively, thereby enhancing model performance. Experimental results on our manually transcribed evaluation set and two public test sets from Common Voice and FLEURS confirm our corpus's high quality and broad applicability. Notably, ASR models trained on GigaSpeech 2 can reduce the word error rate for Thai, Indonesian, and Vietnamese on our challenging and realistic YouTube test set by 25% to 40% compared to Whisper large-v3, with merely 10% model parameters. Furthermore, our ASR models trained on GigaSpeech 2 yield superior performance compared to commercial services. We hope that our newly introduced corpus and pipeline will open a new avenue for low-resource speech recognition and significantly facilitate research in this area.

Zhong Meng, Yu Wu, Naoyuki Kanda et al.

Integrating external language models (LMs) into end-to-end (E2E) models remains a challenging task for domain-adaptive speech recognition. Recently, internal language model estimation (ILME)-based LM fusion has shown significant word error rate (WER) reduction from Shallow Fusion by subtracting a weighted internal LM score from an interpolation of E2E model and external LM scores during beam search. However, on different test sets, the optimal LM interpolation weights vary over a wide range and have to be tuned extensively on well-matched validation sets. In this work, we perform LM fusion in the minimum WER (MWER) training of an E2E model to obviate the need for LM weights tuning during inference. Besides MWER training with Shallow Fusion (MWER-SF), we propose a novel MWER training with ILME (MWER-ILME) where the ILME-based fusion is conducted to generate N-best hypotheses and their posteriors. Additional gradient is induced when internal LM is engaged in MWER-ILME loss computation. During inference, LM weights pre-determined in MWER training enable robust LM integrations on test sets from different domains. Experimented with 30K-hour trained transformer transducers, MWER-ILME achieves on average 8.8% and 5.8% relative WER reductions from MWER and MWER-SF training, respectively, on 6 different test sets

Xie Chen, Sarangarajan Parthasarathy, William Gale et al.

LSTM language models (LSTM-LMs) have been proven to be powerful and yielded significant performance improvements over count based n-gram LMs in modern speech recognition systems. Due to its infinite history states and computational load, most previous studies focus on applying LSTM-LMs in the second-pass for rescoring purpose. Recent work shows that it is feasible and computationally affordable to adopt the LSTM-LMs in the first-pass decoding within a dynamic (or tree based) decoder framework. In this work, the LSTM-LM is composed with a WFST decoder on-the-fly for the first-pass decoding. Furthermore, motivated by the long-term history nature of LSTM-LMs, the use of context beyond the current utterance is explored for the first-pass decoding in conversational speech recognition. The context information is captured by the hidden states of LSTM-LMs across utterance and can be used to guide the first-pass search effectively. The experimental results in our internal meeting transcription system show that significant performance improvements can be obtained by incorporating the contextual information with LSTM-LMs in the first-pass decoding, compared to applying the contextual information in the second-pass rescoring.

Sarangarajan Parthasarathy, William Gale, Xie Chen et al.

We explore neural language modeling for speech recognition where the context spans multiple sentences. Rather than encode history beyond the current sentence using a cache of words or document-level features, we focus our study on the ability of LSTM and Transformer language models to implicitly learn to carry over context across sentence boundaries. We introduce a new architecture that incorporates an attention mechanism into LSTM to combine the benefits of recurrent and attention architectures. We conduct language modeling and speech recognition experiments on the publicly available LibriSpeech corpus. We show that conventional training on a paragraph-level corpus results in significant reductions in perplexity compared to training on a sentence-level corpus. We also describe speech recognition experiments using long-span language models in second-pass re-ranking, and provide insights into the ability of such models to take advantage of context beyond the current sentence.

Yu Wang, Xie Chen, Mark Gales et al.

State-of-the-art English automatic speech recognition systems typically use phonetic rather than graphemic lexicons. Graphemic systems are known to perform less well for English as the mapping from the written form to the spoken form is complicated. However, in recent years the representational power of deep-learning based acoustic models has improved, raising interest in graphemic acoustic models for English, due to the simplicity of generating the lexicon. In this paper, phonetic and graphemic models are compared for an English Multi-Genre Broadcast transcription task. A range of acoustic models based on lattice-free MMI training are constructed using phonetic and graphemic lexicons. For this task, it is found that having a long-span temporal history reduces the difference in performance between the two forms of models. In addition, system combination is examined, using parameter smoothing and hypothesis combination. As the combination approaches become more complicated the difference between the phonetic and graphemic systems further decreases. Finally, for all configurations examined the combination of phonetic and graphemic systems yields consistent gains.