多模型融合的VoxSRC22说话人日志系统

doi:10.3778/j.issn.1002-8331.2301-0080

摘要/Abstract

摘要： 为有效解决“谁在什么时候说话”的问题，提出一种说话人日志方法。该方法由六个模块组成，包括语音活动检测（voice activity detection，VAD）、语音增强、说话人嵌入提取器、说话人聚类、重叠语音检测（overlapping speech detection，OSD）和结果融合。利用语音增强技术可以改善语音活动检测的性能。有效地结合不同的说话人嵌入提取器和聚类算法可以进一步降低系统错误率。在系统融合后处理重叠语音展示了最佳结果。实验结果表明，最佳系统的性能相对基线提升了72%，并在VoxCeleb说话人识别挑战赛（VoxCeleb speaker recognition challenge，VoxSRC）2022评估集上分别实现了5.48%的说话人日志错误率（diarization error rate，DER）和32.10%的杰卡德错误率（Jaccard error rate，JER），排名第四。

关键词: 说话人日志, 语音活动检测, 声纹嵌入, 说话人聚类, 结果融合

Abstract: In order to effectively address the problem of speaker diarization, a novel speaker diarization method is proposed. The proposed method consists of six modules, including voice activity detection (VAD), speech enhancement, speaker embedding extractor, speaker clustering, overlapping speech detection (OSD), and result fusion. The application of speech enhancement techniques can improve the performance of voice activity detection. The effective combination of different speaker embedding extractors and clustering algorithms can further reduce speaker diarization error rate. The best performance is achieved by processing the overlapping speech after system fusion. Experimental results show that the performance of the proposed system outperforms the baseline by 72%, achieves a speaker diarization error rate (DER) of 5.48% and a Jaccard error rate (JER) of 32.10% on the VoxCeleb speaker recognition challenge (VoxSRC) 2022 evaluation set, ranking fourth.

Key words: speaker diarization, voice activity detection, speaker embedding, speaker cluster, result fusion

杜雨轩, 周若华. 多模型融合的VoxSRC22说话人日志系统[J]. 计算机工程与应用, 2024, 60(10): 164-172.

DU Yuxuan, ZHOU Ruohua. Multi-Model Fusion VoxSRC22 Speaker Diarization System[J]. Computer Engineering and Applications, 2024, 60(10): 164-172.

参考文献

[1] ANGUERA X, BOZONNET S, EVANS N, et al. Speaker diarization: a review of recent research[J]. IEEE Transactions on Audio, Speech, and Language Processing, 2012, 20(2): 356-370.
[2] GISH H, SIU M H, ROHLICEK R. Segregation of speakers for speech recognition and speaker identification[C]//Proceedings of the 1991 International Conference on Acoustics, Speech, and Signal Processing, 1991: 873-876.
[3] SIU M H, YU G, GISH H. An unsupervised, sequential learning algorithm for the segmentation of speech waveforms with multiple speakers[C]//Proceedings of the 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing, 1992: 189-192.
[4] ROHLICEK J R, AYUSO D, BATES M, et al. Gisting conversational speech[C]//Proceedings of the 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing, 1992: 113-116.
[5] 凌锦雯，陆伟，刘青松，等. 利用EHMM和CLR的说话人分割聚类算法[J]. 小型微型计算机系统, 2012, 33(6): 1389-1392.
LING J W, LU W, LIU Q S, et al. Speaker diarization using EHMM and CLR[J]. Journal of Chinese Computer Systems, 2012, 33(6): 1389-1392.
[6] CHEN S, GOPALAKRISHNAN P. Speaker, environment and channel change detection and clustering via the Bayesian information criterion[C]//Proceedings of the DARPA Broadcast News Transcription and Understanding Workshop, 1998, 8: 127-132.
[7] CHUNG J S, HUH J, NAGRANI A, et al. Spot the conversation: speaker diarisation in the wild[J]. arXiv:2007.01216, 2020.
[8] 徐志京, 张铁海. 加权全序列卷积神经网络方法的帕金森声纹识别研究[J]. 小型微型计算机系统, 2020, 41(12): 2683-2688.
XU Z J, ZHANG T H. Parkinson voiceprint recognition based on weighted deep full sequence convolutional neural network[J]. Journal of Chinese Computer Systems, 2020, 41(12): 2683-2688.
[9] LANDINI F, GLEMBEK O, MATěJKA P, et al. Analysis of the but diarization system for voxconverse challenge[C]//Proceedings of the 2021 IEEE International Conference on Acoustics, Speech and Signal Processing, 2021: 5819-5823.
[10] VARIANI E, LEI X, MCDERMOTT E, et al. Deep neural networks for small footprint text-dependent speaker verification[C]//Proceedings of the 2014 IEEE International Conference on Acoustics, Speech and Signal Processing, 2014: 4052-4056.
[11] HEIGOLD G, MORENO I, BENGIO S, et al. End-to-end text-dependent speaker verification[C]//Proceedings of the 2016 IEEE International Conference on Acoustics, Speech and Signal Processing, 2016: 5115-5119.
[12] WANG Q, DOWNEY C, WAN L, et al. Speaker diarization with LSTM[C]//Proceedings of the 2018 IEEE International Conference on Acoustics, Speech and Signal Processing, 2018: 5239-5243.
[13] LI Q, KREYSSIG F L, ZHANG C, et al. Discriminative neural clustering for speaker diarisation[C]//Proceedings of the 2021 IEEE Spoken Language Technology Workshop, 2021: 574-581.
[14] PARK T J, HAN K J, KUMAR M, et al. Auto-tuning spectral clustering for speaker diarization using normalized maximum eigengap[J]. IEEE Signal Processing Letters, 2019, 27: 381-385.
[15] KENNY P, REYNOLDS D, CASTALDO F. Diarization of telephone conversations using factor analysis[J]. IEEE Journal of Selected Topics in Signal Processing, 2010, 4(6): 1059-1070.
[16] LANDINI F, PROFANT J, DIEZ M, et al. Bayesian HMM clustering of x-vector sequences (VBx) in speaker diarization: theory, implementation and analysis on standard tasks[J]. Computer Speech & Language, 2022, 71: 101254.
[17] RAJ D, GARCIA-PERERA L P, HUANG Z, et al. DOVER-Lap: a method for combining overlap-aware diarization outputs[C]//Proceedings of the 2021 IEEE Spoken Language Technology Workshop, 2021: 881-888.
[18] RYANT N, SINGH P, KRISHNAMOHAN V, et al. The third DIHARD diarization challenge[J]. arXiv:2012.01477, 2020.
[19] CARLETTA J. Unleashing the killer corpus: experiences in creating the multi-everything AMI meeting corpus[J]. Language Resources and Evaluation, 2007, 41(2): 181-190.
[20] OTTERSON S, OSTENDORF M. Efficient use of overlap information in speaker diarization[C]//Proceedings of the 2007 IEEE Workshop on Automatic Speech Recognition & Understanding, 2007: 683-686.
[21] BREDIN H, YIN R, CORIA J M, et al. Pyannote.Audio: neural building blocks for speaker diarization[C]//Proceedings of the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing, 2020: 7124-7128.
[22] 龙华, 张林濮, 邵玉斌, 等. 说话人特征约束的多任务卷积网络语音增强[J]. 小型微型计算机系统, 2021, 42(10): 2178-2183.
LONG H, ZHANG L P, SHAO Y B, et al. Multi-task convolution network with speaker feature constraint for speech enhancement[J]. Journal of Chinese Computer Systems, 2021, 42(10): 2178-2183.
[23] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[24] NAGRANI A, CHUNG J S, XIE W, et al. VoxCeleb: large-scale speaker verification in the wild[J]. Computer Speech & Language, 2020, 60: 101027.
[25] SNYDER D, CHEN G, POVEY D. MUSAN: a music, speech, and noise corpus[J]. arXiv:1510.08484, 2015.
[26] KO T, PEDDINTI V, POVEY D, et al. A study on data augmentation of reverberant speech for robust speech recognition[C]//Proceedings of the 2017 IEEE International Conference on Acoustics, Speech and Signal Processing, 2017: 5220-5224.
[27] WANG W, QIN X, CHENG M, et al. The DKU-SMIIP diarization system for the VoxCeleb speaker recognition challenge 2022[C]//Proceedings of the VoxSrc Workshop, 2022.
[28] CAI Q, HONG G, YE Z, et al. The Kriston AI system for the VoxCeleb speaker recognition challenge 2022[J]. arXiv:2209.11433, 2022.
[29] PARK D, YU Y, PARK K W, et al. GIST-AiTeR system for the diarization task of the 2022 VoxCeleb speaker recognition challenge[J]. arXiv:2209.10357, 2022.
[30] TEVISSEN Y, BOUDY J, PETITPONT F. The Newsbridge-Telecom SudParis VoxCeleb speaker recognition challenge 2022 system description[J]. arXiv:2301.07491, 2023.
[31] CHOI J H, JEOUNG Y R, KYUNG J, et al. HYU submission for the VoxCeleb speaker recognition challenge 2022[Z]. Hanyang University. Department of Electronic Engineering, 2022.