Qingyang Hong

Hukai Huang, Shenghui Lu, Yahui Shan et al.

The Mixture of Experts (MoE) model is a promising approach for handling code-switching speech recognition (CS-ASR) tasks. However, the existing CS-ASR work on MoE has yet to leverage the advantages of MoE's parameter scaling ability fully. This work proposes DLG-MoE, a Dynamic Language Group-based MoE, which can effectively handle the CS-ASR task and leverage the advantages of parameter scaling. DLG-MoE operates based on a hierarchical routing mechanism. First, the language router explicitly models the language attribute and dispatches the representations to the corresponding language expert groups. Subsequently, the unsupervised router within each language group implicitly models attributes beyond language and coordinates expert routing and collaboration. DLG-MoE outperforms the existing MoE methods on CS-ASR tasks while demonstrating great flexibility. It supports different top-$k$ inference and streaming capabilities and can also prune the model parameters flexibly to obtain a monolingual sub-model. The code has been released.

Hukai Huang, Jiayan Lin, Kaidi Wang et al.

Due to the inherent difficulty in modeling phonetic similarities across different languages, code-switching speech recognition presents a formidable challenge. This study proposes a Collaborative-MoE, a Mixture of Experts (MoE) model that leverages a collaborative mechanism among expert groups. Initially, a preceding routing network explicitly learns Language Identification (LID) tasks and selects experts based on acquired LID weights. This process ensures robust routing information to the MoE layer, mitigating interference from diverse language domains on expert network parameter updates. The LID weights are also employed to facilitate inter-group collaboration, enabling the integration of language-specific representations. Furthermore, within each language expert group, a gating network operates unsupervised to foster collaboration on attributes beyond language. Extensive experiments demonstrate the efficacy of our approach, achieving significant performance enhancements compared to alternative methods. Importantly, our method preserves the efficient inference capabilities characteristic of MoE models without necessitating additional pre-training.

Duojia Li, Shuhan Zhang, Zihan Qian et al.

In target speaker extraction (TSE), we aim to recover target speech from a multi-talker mixture using a short enrollment utterance as reference. Recent studies on diffusion and flow-matching generators have improved target-speech fidelity. However, multi-step sampling increases latency, and one-step solutions often rely on a mixture-dependent time coordinate that can be unreliable for real-world conversations. We present AlphaFlowTSE, a one-step conditional generative model trained with a Jacobian-vector product (JVP)-free AlphaFlow objective. AlphaFlowTSE learns mean-velocity transport along a mixture-to-target trajectory starting from the observed mixture, eliminating auxiliary mixing-ratio prediction, and stabilizes training by combining flow matching with an interval-consistency teacher-student target. Experiments on Libri2Mix and REAL-T confirm that AlphaFlowTSE improves target-speaker similarity and real-mixture generalization for downstream automatic speech recognition (ASR).

Duojia Li, Shenghui Lu, Hongchen Pan et al.

Multistep inference is a bottleneck for real-time generative speech enhancement because flow- and diffusion-based systems learn an instantaneous velocity field and therefore rely on iterative ordinary differential equation (ODE) solvers. We introduce MeanFlowSE, a conditional generative model that learns the average velocity over finite intervals along a trajectory. Using a Jacobian-vector product (JVP) to instantiate the MeanFlow identity, we derive a local training objective that directly supervises finite-interval displacement while remaining consistent with the instantaneous-field constraint on the diagonal. At inference, MeanFlowSE performs single-step generation via a backward-in-time displacement, removing the need for multistep solvers; an optional few-step variant offers additional refinement. On VoiceBank-DEMAND, the single-step model achieves strong intelligibility, fidelity, and perceptual quality with substantially lower computational cost than multistep baselines. The method requires no knowledge distillation or external teachers, providing an efficient, high-fidelity framework for real-time generative speech enhancement. The proposed method is open-sourced at https://github.com/liduojia1/MeanFlowSE.

Kaidi Wang, Wenhao Guan, Ziyue Jiang et al.

Currently, zero-shot voice conversion systems are capable of synthesizing the voice of unseen speakers. However, most existing approaches struggle to accurately replicate the speaking style of the source speaker or mimic the distinctive speaking style of the target speaker, thereby limiting the controllability of voice conversion. In this work, we propose Discl-VC, a novel voice conversion framework that disentangles content and prosody information from self-supervised speech representations and synthesizes the target speaker's voice through in-context learning with a flow matching transformer. To enable precise control over the prosody of generated speech, we introduce a mask generative transformer that predicts discrete prosody tokens in a non-autoregressive manner based on prompts. Experimental results demonstrate the superior performance of Discl-VC in zero-shot voice conversion and its remarkable accuracy in prosody control for synthesized speech.

Kaidi Wang, Wenhao Guan, Shenghui Lu et al.

Recently, flow matching based speech synthesis has significantly enhanced the quality of synthesized speech while reducing the number of inference steps. In this paper, we introduce SlimSpeech, a lightweight and efficient speech synthesis system based on rectified flow. We have built upon the existing speech synthesis method utilizing the rectified flow model, modifying its structure to reduce parameters and serve as a teacher model. By refining the reflow operation, we directly derive a smaller model with a more straight sampling trajectory from the larger model, while utilizing distillation techniques to further enhance the model performance. Experimental results demonstrate that our proposed method, with significantly reduced model parameters, achieves comparable performance to larger models through one-step sampling.

Kaidi Wang, Yi He, Wenhao Guan et al.

Video dubbing aims to generate high-fidelity speech that is precisely temporally aligned with the visual content. Existing methods still suffer from limitations in speech naturalness and audio-visual synchronization, and are limited to monolingual settings. To address these challenges, we propose SyncVoice, a vision-augmented video dubbing framework built upon a pretrained text-to-speech (TTS) model. By fine-tuning the TTS model on audio-visual data, we achieve strong audiovisual consistency. We propose a Dual Speaker Encoder to effectively mitigate inter-language interference in cross-lingual speech synthesis and explore the application of video dubbing in video translation scenarios. Experimental results show that SyncVoice achieves high-fidelity speech generation with strong synchronization performance, demonstrating its potential in video dubbing tasks.

Shenghui Lu, Hukai Huang, Jinanglong Yao et al.

This paper proposes a model that integrates sub-band processing and deep filtering to fully exploit information from the target time-frequency (TF) bin and its surrounding TF bins for single-channel speech enhancement. The sub-band module captures surrounding frequency bin information at the input, while the deep filtering module applies filtering at the output to both the target TF bin and its surrounding TF bins. To further improve the model performance, we decouple deep filtering into temporal and frequency components and introduce a two-stage framework, reducing the complexity of filter coefficient prediction at each stage. Additionally, we propose the TAConv module to strengthen convolutional feature extraction. Experimental results demonstrate that the proposed hierarchical deep filtering network (HDF-Net) effectively utilizes surrounding TF bin information and outperforms other advanced systems while using fewer resources.

Zhaoyang Li, Jie Wang, XiaoXiao Li et al.

In speaker diarization, traditional clustering-based methods remain widely used in real-world applications. However, these methods struggle with the complex distribution of speaker embeddings and overlapping speech segments. To address these limitations, we propose an Overlapping Community Detection method based on Graph Attention networks and the Label Propagation Algorithm (OCDGALP). The proposed framework comprises two key components: (1) a graph attention network that refines speaker embeddings and node connections by aggregating information from neighboring nodes, and (2) a label propagation algorithm that assigns multiple community labels to each node, enabling simultaneous clustering and overlapping community detection. Experimental results show that the proposed method significantly reduces the Diarization Error Rate (DER), achieving a state-of-the-art 15.94% DER on the DIHARD-III dataset without oracle Voice Activity Detection (VAD), and an impressive 11.07% with oracle VAD.

Jie Wang, Yuji Liu, Binling Wang et al.

This paper describes the system developed by the XMUSPEECH team for the Multi-channel Multi-party Meeting Transcription Challenge (M2MeT). For the speaker diarization task, we propose a multi-channel speaker diarization system that obtains spatial information of speaker by Difference of Arrival (DOA) technology. Speaker-spatial embedding is generated by x-vector and s-vector derived from Filter-and-Sum Beamforming (FSB) which makes the embedding more robust. Specifically, we propose a novel multi-channel sequence-to-sequence neural network architecture named Discriminative Multi-stream Neural Network (DMSNet) which consists of Attention Filter-and-Sum block (AFSB) and Conformer encoder. We explore DMSNet to address overlapped speech problem on multi-channel audio. Compared with LSTM based OSD module, we achieve a decreases of 10.1% in Detection Error Rate(DetER). By performing DMSNet based OSD module, the DER of cluster-based diarization system decrease significantly form 13.44% to 7.63%. Our best fusion system achieves 7.09% and 9.80% of the diarization error rate (DER) on evaluation set and test set.

Jie Wang, Fuchuang Tong, Zhicong Chen et al.

This paper describes the XMUSPEECH speaker recognition and diarisation systems for the VoxCeleb Speaker Recognition Challenge 2021. For track 2, we evaluate two systems including ResNet34-SE and ECAPA-TDNN. For track 4, an important part of our system is VAD module which greatly improves the performance. Our best submission on the track 4 obtained on the evaluation set DER 5.54% and JER 27.11%, while the performance on the development set is DER 2.92% and JER 20.84%.

Binling Wang, Wenxuan Hu, Jing Li et al.

This paper introduces the sixth Oriental Language Recognition (OLR) 2021 Challenge, which intends to improve the performance of language recognition systems and speech recognition systems within multilingual scenarios. The data profile, four tasks, two baselines, and the evaluation principles are introduced in this paper. In addition to the Language Identification (LID) tasks, multilingual Automatic Speech Recognition (ASR) tasks are introduced to OLR 2021 Challenge for the first time. The challenge this year focuses on more practical and challenging problems, with four tasks: (1) constrained LID, (2) unconstrained LID, (3) constrained multilingual ASR, (4) unconstrained multilingual ASR. Baselines for LID tasks and multilingual ASR tasks are provided, respectively. The LID baseline system is an extended TDNN x-vector model constructed with Pytorch. A transformer-based end-to-end model is provided as the multilingual ASR baseline system. These recipes will be online published, and available for participants to construct their own LID or ASR systems. The baseline results demonstrate that those tasks are rather challenging and deserve more effort to achieve better performance.

Jing Li, Binling Wang, Yiming Zhi et al.

The fifth Oriental Language Recognition (OLR) Challenge focuses on language recognition in a variety of complex environments to promote its development. The OLR 2020 Challenge includes three tasks: (1) cross-channel language identification, (2) dialect identification, and (3) noisy language identification. We choose Cavg as the principle evaluation metric, and the Equal Error Rate (EER) as the secondary metric. There were 58 teams participating in this challenge and one third of the teams submitted valid results. Compared with the best baseline, the Cavg values of Top 1 system for the three tasks were relatively reduced by 82%, 62% and 48%, respectively. This paper describes the three tasks, the database profile, and the final results. We also outline the novel approaches that improve the performance of language recognition systems most significantly, such as the utilization of auxiliary information.

Dexin Liao, Jing Li, Yiming Zhi et al.

In this paper, we present the XMUSPEECH system for Task 1 of 2020 Personalized Voice Trigger Challenge (PVTC2020). Task 1 is a joint wake-up word detection with speaker verification on close talking data. The whole system consists of a keyword spotting (KWS) sub-system and a speaker verification (SV) sub-system. For the KWS system, we applied a Temporal Depthwise Separable Convolution Residual Network (TDSC-ResNet) to improve the system's performance. For the SV system, we proposed a multi-task learning network, where phonetic branch is trained with the character label of the utterance, and speaker branch is trained with the label of the speaker. Phonetic branch is optimized with connectionist temporal classification (CTC) loss, which is treated as an auxiliary module for speaker branch. Experiments show that our system gets significant improvements compared with baseline system.

Yan Liu, Zheng Li, Lin Li et al.

This paper proposes a multi-task learning network with phoneme-aware and channel-wise attentive learning strategies for text-dependent Speaker Verification (SV). In the proposed structure, the frame-level multi-task learning along with the segment-level adversarial learning is adopted for speaker embedding extraction. The phoneme-aware attentive pooling is exploited on frame-level features in the main network for speaker classifier, with the corresponding posterior probability for the phoneme distribution in the auxiliary subnet. Further, the introduction of Squeeze and Excitation (SE-block) performs dynamic channel-wise feature recalibration, which improves the representational ability. The proposed method exploits speaker idiosyncrasies associated with pass-phrases, and is further improved by the phoneme-aware attentive pooling and SE-block from temporal and channel-wise aspects, respectively. The experiments conducted on RSR2015 Part 1 database confirm that the proposed system achieves outstanding results for textdependent SV.

Zheng Li, Yan Liu, Lin Li et al.

This paper proposes an additive phoneme-aware margin softmax (APM-Softmax) loss to train the multi-task learning network with phonetic information for language recognition. In additive margin softmax (AM-Softmax) loss, the margin is set as a constant during the entire training for all training samples, and that is a suboptimal method since the recognition difficulty varies in training samples. In additive angular margin softmax (AAM-Softmax) loss, the additional angular margin is set as a costant as well. In this paper, we propose an APM-Softmax loss for language recognition with phoneitc multi-task learning, in which the additive phoneme-aware margin is automatically tuned for different training samples. More specifically, the margin of language recognition is adjusted according to the results of phoneme recognition. Experiments are reported on Oriental Language Recognition (OLR) datasets, and the proposed method improves AM-Softmax loss and AAM-Softmax loss in different language recognition testing conditions.

Zheng Li, Miao Zhao, Qingyang Hong et al.

This paper introduces the fifth oriental language recognition (OLR) challenge AP20-OLR, which intends to improve the performance of language recognition systems, along with APSIPA Annual Summit and Conference (APSIPA ASC). The data profile, three tasks, the corresponding baselines, and the evaluation principles are introduced in this paper. The AP20-OLR challenge includes more languages, dialects and real-life data provided by Speechocean and the NSFC M2ASR project, and all the data is free for participants. The challenge this year still focuses on practical and challenging problems, with three tasks: (1) cross-channel LID, (2) dialect identification and (3) noisy LID. Based on Kaldi and Pytorch, recipes for i-vector and x-vector systems are also conducted as baselines for the three tasks. These recipes will be online-published, and available for participants to configure LID systems. The baseline results on the three tasks demonstrate that those tasks in this challenge are worth paying more efforts to achieve better performance.

Jianfeng Zhou, Tao Jiang, Lin Li et al.

Under noisy environments, to achieve the robust performance of speaker recognition is still a challenging task. Motivated by the promising performance of multi-task training in a variety of image processing tasks, we explore the potential of multi-task adversarial training for learning a noise-robust speaker embedding. In this paper we present a novel framework which consists of three components: an encoder that extracts noise-robust speaker embedding; a classifier that classifies the speakers; a discriminator that discriminates the noise type of the speaker embedding. Besides, we propose a training strategy using the training accuracy as an indicator to stabilize the multi-class adversarial optimization process. We conduct our experiments on the English and Mandarin corpus and the experimental results demonstrate that our proposed multi-task adversarial training method could greatly outperform the other methods without adversarial training in noisy environments. Furthermore, experiments indicate that our method is also able to improve the speaker verification performance the clean condition.