Tian-hao Zhang

Yuwen Wang, Xinyuan Qian, Tian-Hao Zhang et al.

Large Audio-Language Models (LALMs) have demonstrated strong performance in audio understanding and generation. Yet, our extensive benchmarking reveals that their behavior is largely generic (e.g., summarizing spoken content) and fails to adequately support personalized question answering (e.g., summarizing what my best friend says). In contrast, human conditions their interpretation and decision-making on each individual's personal context. To bridge this gap, we formalize the task of Personalized LALMs (PALM) for recognizing personal concepts and reasoning within personal context. Moreover, we create the first benchmark (PALM-Bench) to foster the methodological advances in PALM and enable structured evaluation on several tasks across multi-speaker scenarios. Our extensive experiments on representative open-source LALMs, show that existing training-free prompting and supervised fine-tuning strategies, while yield improvements, remains limited in modeling personalized knowledge and transferring them across tasks robustly. Data and code will be released.

Tian-Hao Zhang, Dinghao Zhou, Guiping Zhong et al.

RNN-T models are widely used in ASR, which rely on the RNN-T loss to achieve length alignment between input audio and target sequence. However, the implementation complexity and the alignment-based optimization target of RNN-T loss lead to computational redundancy and a reduced role for predictor network, respectively. In this paper, we propose a novel model named CIF-Transducer (CIF-T) which incorporates the Continuous Integrate-and-Fire (CIF) mechanism with the RNN-T model to achieve efficient alignment. In this way, the RNN-T loss is abandoned, thus bringing a computational reduction and allowing the predictor network a more significant role. We also introduce Funnel-CIF, Context Blocks, Unified Gating and Bilinear Pooling joint network, and auxiliary training strategy to further improve performance. Experiments on the 178-hour AISHELL-1 and 10000-hour WenetSpeech datasets show that CIF-T achieves state-of-the-art results with lower computational overhead compared to RNN-T models.

Chunyu Sun, Bingyu Liu, Zhichao Cui et al.

Embedding-based retrieval models have made significant strides in retrieval-augmented generation (RAG) techniques for text and multimodal large language models (LLMs) applications. However, when it comes to speech larage language models (SLLMs), these methods are limited to a two-stage process, where automatic speech recognition (ASR) is combined with text-based retrieval. This sequential architecture suffers from high latency and error propagation. To address these limitations, we propose a unified embedding framework that eliminates the need for intermediate text representations. Specifically, the framework includes separate speech and text encoders, followed by a shared scaling layer that maps both modalities into a common embedding space. Our model reduces pipeline latency by 50\% while achieving higher retrieval accuracy compared to traditional two-stage methods. We also provide a theoretical analysis of the challenges inherent in end-to-end speech retrieval and introduce architectural principles for effective speech-to-document matching. Extensive experiments demonstrate the robustness of our approach across diverse acoustic conditions and speaker variations, paving the way for a new paradigm in multimodal SLLMs retrieval systems.

Tian-Hao Zhang, Jiawei Zhang, Jun Wang et al.

Humans can perceive speakers' characteristics (e.g., identity, gender, personality and emotion) by their appearance, which are generally aligned to their voice style. Recently, vision-driven Text-to-speech (TTS) scholars grounded their investigations on real-person faces, thereby restricting effective speech synthesis from applying to vast potential usage scenarios with diverse characters and image styles. To solve this issue, we introduce a novel FaceSpeak approach. It extracts salient identity characteristics and emotional representations from a wide variety of image styles. Meanwhile, it mitigates the extraneous information (e.g., background, clothing, and hair color, etc.), resulting in synthesized speech closely aligned with a character's persona. Furthermore, to overcome the scarcity of multi-modal TTS data, we have devised an innovative dataset, namely Expressive Multi-Modal TTS, which is diligently curated and annotated to facilitate research in this domain. The experimental results demonstrate our proposed FaceSpeak can generate portrait-aligned voice with satisfactory naturalness and quality.

Wei Zhang, Tian-Hao Zhang, Chao Luo et al.

Recently, end-to-end automatic speech recognition has become the mainstream approach in both industry and academia. To optimize system performance in specific scenarios, the Weighted Finite-State Transducer (WFST) is extensively used to integrate acoustic and language models, leveraging its capacity to implicitly fuse language models within static graphs, thereby ensuring robust recognition while also facilitating rapid error correction. However, WFST necessitates a frame-by-frame search of CTC posterior probabilities through autoregression, which significantly hampers inference speed. In this work, we thoroughly investigate the spike property of CTC outputs and further propose the conjecture that adjacent frames to non-blank spikes carry semantic information beneficial to the model. Building on this, we propose the Spike Window Decoding algorithm, which greatly improves the inference speed by making the number of frames decoded in WFST linearly related to the number of spiking frames in the CTC output, while guaranteeing the recognition performance. Our method achieves SOTA recognition accuracy with significantly accelerates decoding speed, proven across both AISHELL-1 and large-scale In-House datasets, establishing a pioneering approach for integrating CTC output with WFST.

Jiaming Zhou, Shiwan Zhao, Hui Wang et al.

The kNN-CTC model has proven to be effective for monolingual automatic speech recognition (ASR). However, its direct application to multilingual scenarios like code-switching, presents challenges. Although there is potential for performance improvement, a kNN-CTC model utilizing a single bilingual datastore can inadvertently introduce undesirable noise from the alternative language. To address this, we propose a novel kNN-CTC-based code-switching ASR (CS-ASR) framework that employs dual monolingual datastores and a gated datastore selection mechanism to reduce noise interference. Our method selects the appropriate datastore for decoding each frame, ensuring the injection of language-specific information into the ASR process. We apply this framework to cutting-edge CTC-based models, developing an advanced CS-ASR system. Extensive experiments demonstrate the remarkable effectiveness of our gated datastore mechanism in enhancing the performance of zero-shot Chinese-English CS-ASR.

Zhi-Hao Lai, Tian-Hao Zhang, Qi Liu et al.

The local and global features are both essential for automatic speech recognition (ASR). Many recent methods have verified that simply combining local and global features can further promote ASR performance. However, these methods pay less attention to the interaction of local and global features, and their series architectures are rigid to reflect local and global relationships. To address these issues, this paper proposes InterFormer for interactive local and global features fusion to learn a better representation for ASR. Specifically, we combine the convolution block with the transformer block in a parallel design. Besides, we propose a bidirectional feature interaction module (BFIM) and a selective fusion module (SFM) to implement the interaction and fusion of local and global features, respectively. Extensive experiments on public ASR datasets demonstrate the effectiveness of our proposed InterFormer and its superior performance over the other Transformer and Conformer models.

Tian-Hao Zhang, Hai-Bo Qin, Zhi-Hao Lai et al.

Attention-based encoder-decoder (AED) models have shown impressive performance in ASR. However, most existing AED methods neglect to simultaneously leverage both acoustic and semantic features in decoder, which is crucial for generating more accurate and informative semantic states. In this paper, we propose an Acoustic and Semantic Cooperative Decoder (ASCD) for ASR. In particular, unlike vanilla decoders that process acoustic and semantic features in two separate stages, ASCD integrates them cooperatively. To prevent information leakage during training, we design a Causal Multimodal Mask. Moreover, a variant Semi-ASCD is proposed to balance accuracy and computational cost. Our proposal is evaluated on the publicly available AISHELL-1 and aidatatang_200zh datasets using Transformer, Conformer, and Branchformer as encoders, respectively. The experimental results show that ASCD significantly improves the performance by leveraging both the acoustic and semantic information cooperatively.

Chuan-Fei Zhang, Yan Liu, Tian-Hao Zhang et al.

Non-autoregressive (NAR) transformer models have been studied intensively in automatic speech recognition (ASR), and a substantial part of NAR transformer models is to use the casual mask to limit token dependencies. However, the casual mask is designed for the left-to-right decoding process of the non-parallel autoregressive (AR) transformer, which is inappropriate for the parallel NAR transformer since it ignores the right-to-left contexts. Some models are proposed to utilize right-to-left contexts with an extra decoder, but these methods increase the model complexity. To tackle the above problems, we propose a new non-autoregressive transformer with a unified bidirectional decoder (NAT-UBD), which can simultaneously utilize left-to-right and right-to-left contexts. However, direct use of bidirectional contexts will cause information leakage, which means the decoder output can be affected by the character information from the input of the same position. To avoid information leakage, we propose a novel attention mask and modify vanilla queries, keys, and values matrices for NAT-UBD. Experimental results verify that NAT-UBD can achieve character error rates (CERs) of 5.0%/5.5% on the Aishell1 dev/test sets, outperforming all previous NAR transformer models. Moreover, NAT-UBD can run 49.8x faster than the AR transformer baseline when decoding in a single step.