Keyu An

Keyu An, Qian Chen, Chong Deng et al.

This report introduces FunAudioLLM, a model family designed to enhance natural voice interactions between humans and large language models (LLMs). At its core are two innovative models: SenseVoice, which handles multilingual speech recognition, emotion recognition, and audio event detection; and CosyVoice, which facilitates natural speech generation with control over multiple languages, timbre, speaking style, and speaker identity. SenseVoice-Small delivers exceptionally low-latency ASR for 5 languages, and SenseVoice-Large supports high-precision ASR for over 50 languages, while CosyVoice excels in multi-lingual voice generation, zero-shot in-context learning, cross-lingual voice cloning, and instruction-following capabilities. The models related to SenseVoice and CosyVoice have been open-sourced on Modelscope and Huggingface, along with the corresponding training, inference, and fine-tuning codes released on GitHub. By integrating these models with LLMs, FunAudioLLM enables applications such as speech-to-speech translation, emotional voice chat, interactive podcasts, and expressive audiobook narration, thereby pushing the boundaries of voice interaction technology. Demos are available at https://fun-audio-llm.github.io, and the code can be accessed at https://github.com/FunAudioLLM.

Keyu An, Shiliang Zhang, Zhijie Yan

In this study, we delve into the efficacy of transformers within pre-trained language models (PLMs) when repurposed as encoders for Automatic Speech Recognition (ASR). Our underlying hypothesis posits that, despite being initially trained on text-based corpora, these transformers possess a remarkable capacity to extract effective features from the input sequence. This inherent capability, we argue, is transferrable to speech data, thereby augmenting the acoustic modeling ability of ASR. Through rigorous empirical analysis, our findings reveal a notable improvement in Character Error Rate (CER) and Word Error Rate (WER) across diverse ASR tasks when transformers from pre-trained LMs are incorporated. Particularly, they serve as an advantageous starting point for initializing ASR encoders. Furthermore, we uncover that these transformers, when integrated into a well-established ASR encoder, can significantly boost performance, especially in scenarios where profound semantic comprehension is pivotal. This underscores the potential of leveraging the semantic prowess embedded within pre-trained transformers to advance ASR systems' capabilities.

Cong Wang, Changfeng Gao, Yang Xiang et al.

Differentiable reinforcement learning (RL) frameworks like DiffRO offer a powerful approach for controllable text-to-speech (TTS), but are vulnerable to reward hacking, particularly for nuanced tasks like emotion control. The policy model can exploit a vanilla Reward Model (RM) by generating acoustic artifacts to achieve spurious rewards, but at the cost of degrading perceptual quality. To address this, we propose Robust Reward Policy Optimization (RRPO), a novel framework that employs a hybrid regularization scheme. This scheme develops a robust RM whose reward signal is more reliably aligned with human perception, compelling the policy to abandon detrimental shortcuts and instead learn the complex features of genuine emotions. Our ablation study confirms the enhanced robustness of our RM, as evidenced by its strong cross-lingual generalization. The subjective evaluation demonstrates that this robust RM effectively mitigates reward hacking, leading to significant improvements in both emotional expressiveness and naturalness over all baselines. Demo page: https://lrwinr.github.io/RRPO-CosyVoice.

Keyu An, Yanni Chen, Chong Deng et al.

In recent years, automatic speech recognition (ASR) has witnessed transformative advancements driven by three complementary paradigms: data scaling, model size scaling, and deep integration with large language models (LLMs). However, LLMs are prone to hallucination, which can significantly degrade user experience in real-world ASR applications. In this paper, we present Fun-ASR, a large-scale, LLM-based ASR system that synergistically combines massive data, large model capacity, LLM integration, and reinforcement learning to achieve state-of-the-art performance across diverse and complex speech recognition scenarios. Moreover, Fun-ASR is specifically optimized for practical deployment, with enhancements in streaming capability, noise robustness, code-switching, hotword customization, and satisfying other real-world application requirements. Experimental results show that while most LLM-based ASR systems achieve strong performance on open-source benchmarks, they often underperform on real industry evaluation sets. Thanks to production-oriented optimizations, Fun-ASR achieves state-of-the-art performance on real application datasets, demonstrating its effectiveness and robustness in practical settings.

Keyu An, Hongyu Xiang, Zhijian Ou

In this paper, we present a new open source toolkit for speech recognition, named CAT (CTC-CRF based ASR Toolkit). CAT inherits the data-efficiency of the hybrid approach and the simplicity of the E2E approach, providing a full-fledged implementation of CTC-CRFs and complete training and testing scripts for a number of English and Chinese benchmarks. Experiments show CAT obtains state-of-the-art results, which are comparable to the fine-tuned hybrid models in Kaldi but with a much simpler training pipeline. Compared to existing non-modularized E2E models, CAT performs better on limited-scale datasets, demonstrating its data efficiency. Furthermore, we propose a new method called contextualized soft forgetting, which enables CAT to do streaming ASR without accuracy degradation. We hope CAT, especially the CTC-CRF based framework and software, will be of broad interest to the community, and can be further explored and improved.

Keyu An, Hongyu Xiang, Zhijian Ou

In this paper, we present a new open source toolkit for automatic speech recognition (ASR), named CAT (CRF-based ASR Toolkit). A key feature of CAT is discriminative training in the framework of conditional random field (CRF), particularly with connectionist temporal classification (CTC) inspired state topology. CAT contains a full-fledged implementation of CTC-CRF and provides a complete workflow for CRF-based end-to-end speech recognition. Evaluation results on Chinese and English benchmarks such as Switchboard and Aishell show that CAT obtains the state-of-the-art results among existing end-to-end models with less parameters, and is competitive compared with the hybrid DNN-HMM models. Towards flexibility, we show that i-vector based speaker-adapted recognition and latency control mechanism can be explored easily and effectively in CAT. We hope CAT, especially the CRF-based framework and software, will be of broad interest to the community, and can be further explored and improved.

Zhihao Du, Changfeng Gao, Yuxuan Wang et al.

In our prior works, we introduced a scalable streaming speech synthesis model, CosyVoice 2, which integrates a large language model (LLM) and a chunk-aware flow matching (FM) model, and achieves low-latency bi-streaming speech synthesis and human-parity quality. Despite these advancements, CosyVoice 2 exhibits limitations in language coverage, domain diversity, data volume, text formats, and post-training techniques. In this paper, we present CosyVoice 3, an improved model designed for zero-shot multilingual speech synthesis in the wild, surpassing its predecessor in content consistency, speaker similarity, and prosody naturalness. Key features of CosyVoice 3 include: 1) A novel speech tokenizer to improve prosody naturalness, developed via supervised multi-task training, including automatic speech recognition, speech emotion recognition, language identification, audio event detection, and speaker analysis. 2) A new differentiable reward model for post-training applicable not only to CosyVoice 3 but also to other LLM-based speech synthesis models. 3) Dataset Size Scaling: Training data is expanded from ten thousand hours to one million hours, encompassing 9 languages and 18 Chinese dialects across various domains and text formats. 4) Model Size Scaling: Model parameters are increased from 0.5 billion to 1.5 billion, resulting in enhanced performance on our multilingual benchmark due to the larger model capacity. These advancements contribute significantly to the progress of speech synthesis in the wild. We encourage readers to listen to the demo at https://funaudiollm.github.io/cosyvoice3.

Changfeng Gao, Yabin Li, Keyu An et al.

In recent years, large language models (LLMs) have played an important role in automatic speech recognition (ASR) and text-to-speech (TTS) systems. While reinforcement learning (RL) has significantly enhanced LLM performance in text-based tasks, its application to ASR and TTS remains underexplored due to the complexity of training audio-based models. In this study, we propose a lightweight RL framework tailored for audio-based LLMs that can process audio inputs and generate audio outputs. Based on this framework, we evaluate the effectiveness of reinforcement learning on both ASR and TTS tasks. For the ASR task, we experiment with different rule-based reward functions within the Group Relative Policy Optimization (GRPO) framework and investigate the impact of RL data construction. For the TTS task, we compare GRPO with Differentiable Reward Optimization (DiffRO) and further combine the two approaches to achieve improved performance. Our experiments demonstrate that RL can significantly enhance the performance of both ASR and TTS systems, even with limited training data and a small number of optimization steps.

Huahuan Zheng, Keyu An, Zhijian Ou et al.

Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-T) for speech recognition is challenging. Recently, a class of methods such as density ratio (DR) and internal language model estimation (ILME) have been developed, outperforming the classic shallow fusion (SF) method. The basic idea behind these methods is that RNN-T posterior should first subtract the implicitly learned internal language model (ILM) prior, in order to integrate the ELM. While recent studies suggest that RNN-T only learns some low-order language model information, the DR method uses a well-trained neural language model with full context, which may be inappropriate for the estimation of ILM and deteriorate the integration performance. Based on the DR method, we propose a low-order density ratio method (LODR) by replacing the estimation with a low-order weak language model. Extensive empirical experiments are conducted on both in-domain and cross-domain scenarios on English LibriSpeech & Tedlium-2 and Chinese WenetSpeech & AISHELL-1 datasets. It is shown that LODR consistently outperforms SF in all tasks, while performing generally close to ILME and better than DR in most tests.

Keyu An, Huahuan Zheng, Zhijian Ou et al.

History and future contextual information are known to be important for accurate acoustic modeling. However, acquiring future context brings latency for streaming ASR. In this paper, we propose a new framework - Chunking, Simulating Future Context and Decoding (CUSIDE) for streaming speech recognition. A new simulation module is introduced to recursively simulate the future contextual frames, without waiting for future context. The simulation module is jointly trained with the ASR model using a self-supervised loss; the ASR model is optimized with the usual ASR loss, e.g., CTC-CRF as used in our experiments. Experiments show that, compared to using real future frames as right context, using simulated future context can drastically reduce latency while maintaining recognition accuracy. With CUSIDE, we obtain new state-of-the-art streaming ASR results on the AISHELL-1 dataset.

Keyu An, Ji Xiao, Zhijian Ou

Recently, the end-to-end training approach for multi-channel ASR has shown its effectiveness, which usually consists of a beamforming front-end and a recognition back-end. However, the end-to-end training becomes more difficult due to the integration of multiple modules, particularly considering that multi-channel speech data recorded in real environments are limited in size. This raises the demand to exploit the single-channel data for multi-channel end-to-end ASR. In this paper, we systematically compare the performance of three schemes to exploit external single-channel data for multi-channel end-to-end ASR, namely back-end pre-training, data scheduling, and data simulation, under different settings such as the sizes of the single-channel data and the choices of the front-end. Extensive experiments on CHiME-4 and AISHELL-4 datasets demonstrate that while all three methods improve the multi-channel end-to-end speech recognition performance, data simulation outperforms the other two, at the cost of longer training time. Data scheduling outperforms back-end pre-training marginally but nearly consistently, presumably because that in the pre-training stage, the back-end tends to overfit on the single-channel data, especially when the single-channel data size is small.

Chengrui Zhu, Keyu An, Huahuan Zheng et al.

The use of phonological features (PFs) potentially allows language-specific phones to remain linked in training, which is highly desirable for information sharing for multilingual and crosslingual speech recognition methods for low-resourced languages. A drawback suffered by previous methods in using phonological features is that the acoustic-to-PF extraction in a bottom-up way is itself difficult. In this paper, we propose to join phonology driven phone embedding (top-down) and deep neural network (DNN) based acoustic feature extraction (bottom-up) to calculate phone probabilities. The new method is called JoinAP (Joining of Acoustics and Phonology). Remarkably, no inversion from acoustics to phonological features is required for speech recognition. For each phone in the IPA (International Phonetic Alphabet) table, we encode its phonological features to a phonological-vector, and then apply linear or nonlinear transformation of the phonological-vector to obtain the phone embedding. A series of multilingual and crosslingual (both zero-shot and few-shot) speech recognition experiments are conducted on the CommonVoice dataset (German, French, Spanish and Italian) and the AISHLL-1 dataset (Mandarin), and demonstrate the superiority of JoinAP with nonlinear phone embeddings over both JoinAP with linear phone embeddings and the traditional method with flat phone embeddings.

Keyu An, Yi Zhang, Zhijian Ou

Time Delay Neural Networks (TDNNs) are widely used in both DNN-HMM based hybrid speech recognition systems and recent end-to-end systems. Nevertheless, the receptive fields of TDNNs are limited and fixed, which is not desirable for tasks like speech recognition, where the temporal dynamics of speech are varied and affected by many factors. This paper proposes to use deformable TDNNs for adaptive temporal dynamics modeling in end-to-end speech recognition. Inspired by deformable ConvNets, deformable TDNNs augment the temporal sampling locations with additional offsets and learn the offsets automatically based on the ASR criterion, without additional supervision. Experiments show that deformable TDNNs obtain state-of-the-art results on WSJ benchmarks (1.42\%/3.45\% WER on WSJ eval92/dev93 respectively), outperforming standard TDNNs significantly. Furthermore, we propose the latency control mechanism for deformable TDNNs, which enables deformable TDNNs to do streaming ASR without accuracy degradation.

Fan Yu, Zhuoyuan Yao, Xiong Wang et al.

Automatic speech recognition (ASR) has been significantly advanced with the use of deep learning and big data. However improving robustness, including achieving equally good performance on diverse speakers and accents, is still a challenging problem. In particular, the performance of children speech recognition (CSR) still lags behind due to 1) the speech and language characteristics of children's voice are substantially different from those of adults and 2) sizable open dataset for children speech is still not available in the research community. To address these problems, we launch the Children Speech Recognition Challenge (CSRC), as a flagship satellite event of IEEE SLT 2021 workshop. The challenge will release about 400 hours of Mandarin speech data for registered teams and set up two challenge tracks and provide a common testbed to benchmark the CSR performance. In this paper, we introduce the datasets, rules, evaluation method as well as baselines.

Huahuan Zheng, Keyu An, Zhijian Ou

Neural Architecture Search (NAS), the process of automating architecture engineering, is an appealing next step to advancing end-to-end Automatic Speech Recognition (ASR), replacing expert-designed networks with learned, task-specific architectures. In contrast to early computational-demanding NAS methods, recent gradient-based NAS methods, e.g., DARTS (Differentiable ARchiTecture Search), SNAS (Stochastic NAS) and ProxylessNAS, significantly improve the NAS efficiency. In this paper, we make two contributions. First, we rigorously develop an efficient NAS method via Straight-Through (ST) gradients, called ST-NAS. Basically, ST-NAS uses the loss from SNAS but uses ST to back-propagate gradients through discrete variables to optimize the loss, which is not revealed in ProxylessNAS. Using ST gradients to support sub-graph sampling is a core element to achieve efficient NAS beyond DARTS and SNAS. Second, we successfully apply ST-NAS to end-to-end ASR. Experiments over the widely benchmarked 80-hour WSJ and 300-hour Switchboard datasets show that the ST-NAS induced architectures significantly outperform the human-designed architecture across the two datasets. Strengths of ST-NAS such as architecture transferability and low computation cost in memory and time are also reported.