语音识别中说话人自适应方法研究综述

doi:10.3778/j.issn.1673-9418.2104068

摘要/Abstract

摘要：

语音是人机交互方式之一，语音识别技术是人工智能的重要组成部分。近年来神经网络技术在语音识别领域的应用快速发展，已经成为语音识别领域中主流的声学建模技术。然而测试条件中目标说话人语音与训练数据存在差异，导致模型不适配的问题。因此说话人自适应（SA）方法是为了解决说话人差异导致的不匹配问题，研究说话人自适应方法成为语音识别领域的一个热门方向。相比传统语音识别模型中的说话人自适应方法，使用神经网络的语音识别系统中的自适应存在着模型参数庞大，而自适应数据量相对较少等特点，这使得基于神经网络的语音识别系统中的说话人自适应方法成为一个研究难题。首先回顾说话人自适应方法的发展历程和基于神经网络的说话人自适应方法研究遇到的各种问题，其次将说话人自适应方法分为基于特征域和基于模型域的说话人自适应方法并介绍对应原理和改进方法，最后指出说话人自适应方法在语音识别中仍然存在的问题及未来的发展方向。

关键词: 语音识别, 说话人自适应（SA）, 神经网络

Abstract:

Speech is one of the ways of human-computer interaction, and speech recognition technology is an important part of artificial intelligence. In recent years, the application of neural network technology in the field of speech recognition has developed rapidly, and it has become the mainstream acoustic modeling technology in the field of speech recognition. However, there is a difference between target speaker??s voice and training data in the test conditions, which leads to the problem of model incompatibility. Therefore, the speaker adaptation (SA) method is to solve the mismatch problem caused by the speaker difference, and the research on the speaker adaptation method has become a popular direction in the field of speech recognition. Compared with the speaker adaptation method in the traditional speech recognition system, the self-adaptation in the speech recognition system using neural network has the characteristics of huge model parameters and relatively small amount of data. Therefore, the speaker adaptation method in the neural network-based speech recognition system becomes a challenge. Firstly, this paper reviews the development history of the speaker adaptation method and the various problems encountered in the research of the neural network-based speaker adaptation method. Secondly, the speaker adaptation method is divided into the speaker adaptation method based on feature domain and the speaker adaptation method based on model domain. It also introduces the corresponding principles and improvement methods, and finally points out the pro-blems that still exist in the speaker adaptation method in speech recognition and the future development direction.

Key words: speech recognition, speaker adaptation (SA), neural network

朱方圆, 马志强, 陈艳, 张晓旭, 王洪彬, 宝财吉拉呼. 语音识别中说话人自适应方法研究综述[J]. 计算机科学与探索, 2021, 15(12): 2241-2255.

ZHU Fangyuan, MA Zhiqiang, CHEN Yan, ZHANG Xiaoxu, WANG Hongbin, BAO Caijilahu. Survey of Speaker Adaptation Methods in Speech Recognition[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(12): 2241-2255.

参考文献

[1] YU D, DENG L, YU K, et al. Analyzing deep learning: speech recognition practice[M]. Beijing: Electronic Industry Press, 2016.
俞栋, 邓力, 俞凯, 等. 解析深度学习: 语音识别实践[M]. 北京：电子工业出版社, 2016.
[2] GALES M J F, WOODLAND P C. Mean and variance adap-tation within the MLLR framework[J]. Computer Speech & Language, 1996, 10(4): 249-264.
[3] GALES M J F, PYE D, WOODLAND P C. Variance com-pensation within the MLLR framework for robust speech recognition and speaker adaptation[C]//Proceedings of the 4th International Conference on Spoken Language Processing, Philadelphia, Oct 3-6, 1996: 1832-1835.
[4] NGUYEN P, WELLEKENS C, JUNQUA J C. Maximum likelihood eigenspace and MLLR for speech recognition in noisy environments[C]//Proceedings of the 6th European Conference on Speech Communication and Technology, Bu-dapest, Sep 5-9, 1999: 2519-2522.
[5] VARADARAJAN B, POVEY D, CHU S M. Quick fMLLR for speaker adaptation in speech recognition[C]//Proceedings of the 2008 IEEE International Conference on Acoustics, Speech, and Signal Processing, Caesars Palace, Mar 30-Apr 4, 2008. Piscataway: IEEE, 2008: 4297-4300.
[6] JAFARI K, ALMASGANJ F, SHEKOFTEH Y. Combination of fMLLR with clustering and fMLLR with MLLR clustering for rapid speaker adaptation[C]//Proceedings of the 2010 2nd International Conference on Electronic Computer Technology, Kuala Lumpur, May 7-10, 2010. Piscataway: IEEE, 2010: 133-136.
[7] WANG D, NARAYANAN S S. A confidence-score based un-supervised MAP adaptation for speech recognition[C]//Pro-ceedings of the 2002 Conference Record of the 36th Asilomar Conference on Signals, Systems and Computers, Pacific Grove, Nov 3-6, 2002. Piscataway: IEEE, 2002: 222-226.
[8] CHESTA C, SIOHAN O, LEE C H. Maximum a posteriori linear regression for hidden Markov model adaptation[C]//Proceedings of the 6th European Conference on Speech Communication and Technology, Budapest, Sep 5-9, 1999: 211-214.
[9] LEE L, ROSE R C. A frequency warping approach to speaker normalization[J]. IEEE Transactions on Speech and Audio Processing, 1998, 6(1): 49-60.
[10] SANAND D R, KUMAR D D, UMESH S. Linear trans-formation approach to VTLN using dynamic frequency warping[C]//Proceedings of the 8th Annual Conference of the International Speech Communication Association, Antwerp, Aug 27-31, 2007: 1138-1141.
[11] CUI X D, ALWAN A. MLLR-like speaker adaptation based on linearization of VTLN with MFCC features[C]//Procee-dings of the 9th European Conference on Speech Communi-cation and Technology, Lisbon, Sep 4-8, 2005: 273-276.
[12] NETO J P, ALMEIDA L B, HOCHBERG M, et al. Speaker-adaptation for hybrid HMM-ANN continuous speech reco-gnition system[C]//Proceedings of the 4th European Confer-ence on Speech Communication and Technology, Madrid, Sep 18-21, 1995: 2171-2174.
[13] PAN J. Research on adaptive methods in deep learning speech recognition system[D]. Hefei: University of Science and Technology of China, 2020.
潘嘉. 深度学习语音识别系统中的自适应方法研究[D]. 合肥: 中国科学技术大学, 2020.
[14] GEMELLO R, MANA F, SCANZIO S, et al. Adaptation of hybrid ANN/HMM models using linear hidden transforma-tions and conservative training[C]//Proceedings of the 2006 IEEE International Conference on Acoustics Speech and Signal Processing, Toulouse, May 14-19, 2006. Piscataway: IEEE, 2006: 1189-1192.
[15] LI B, SIM K C. Comparison of discriminative input and output transformations for speaker adaptation in the hybrid NN/HMM systems[C]//Proceedings of the 11th Annual Con-ference of the International Speech Communication Associa-tion, Makuhari, Sep 26-30, 2010: 526-529.
[16] TRMAL J, ZELINKA J, MüLLER L. Adaptation of a feed-forward artificial neural network using a linear transform[C]//LNCS 6231: Proceedings of the 13th International Conference on Text, Speech and Dialogue, Brno, Sep 6-10, 2010. Berlin, Heidelberg: Springer, 2010: 423-430.
[17] XIAO Y M, ZHANG Z, CAI S, et al. A initial attempt on task-specific adaptation for deep neural network-based large vocabulary continuous speech recognition[C]//Proceedings of the 13th Annual Conference of the International Speech Communication Association, Portland, Sep 9-13, 2012: 2574-2577.
[18] YAO K S, DONG Y, SEIDE F, et al. Adaptation of context-dependent deep neural networks for automatic speech reco-gnition[C]//Proceedings of the 2012 IEEE Spoken Language Technology Workshop, Miami, Dec 2-5, 2012. Piscataway: IEEE, 2012: 366-369.
[19] KARTHICK B M, KOLHAR P, UMESH S. Speaker adapta-tion of convolutional neural network using speaker specific subspace vectors of SGMM[C]//Proceedings of the 16th Annual Conference of the International Speech Communi-cation Association, Dresden, Sep 6-10, 2015: 1096-1100.
[20] YI J Y, TAO J H. Batch normalization based unsupervised speaker adaptation for acoustic models[C]//Proceedings of the 2019 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, Lanzhou, Nov 18-21, 2019. Piscataway: IEEE, 2019: 176-180.
[21] TOMASHENKO N A, ESTèVE Y. Evaluation of feature-space speaker adaptation for end-to-end acoustic models[C]//Proceedings of the 11th International Conference on Language Resources and Evaluation, Miyazaki, May 7-12, 2018.
[22] SAON G, SOLTAU H, NAHAMOO D, et al. Speaker ada-ptation of neural network acoustic models using i-vectors[C]//Proceedings of the 2013 IEEE Workshop on Automatic Speech Recognition and Understanding, Olomouc, Dec 8-12, 2013. Piscataway: IEEE, 2013: 55-59.
[23] KARANASOU P, WANG Y Q, GALES M J F, et al. Adaptation of deep neural network acoustic models using factorised i-vectors[C]//Proceedings of the 15th Annual Con-ference of the International Speech Communication Asso-ciation, Singapore, Sep 14-18, 2014: 2180-2184.
[24] MIAO Y J, ZHANG H, METZE F. Speaker adaptive training of deep neural network acoustic models using i-vectors[J]. IEEE Transactions on Audio, Speech and Language Pro-cessing, 2015, 23(11): 1938-1949.
[25] CARDINAL P, DEHAK N, ZHANG Y, et al. Speaker ada-ptation using the i-vector technique for bottleneck features[C]//Proceedings of the 16th Annual Conference of the International Speech Communication Association, Dresden, Sep 6-10, 2015: 2867-2871.
[26] CUI X D, GOEL V, SAON G, et al. Embedding-based speaker adaptive training of deep neural networks[C]//Proceedings of the 18th Annual Conference of the International Speech Communication Association, Stockholm, Aug 20-24, 2017: 122-126.
[27] JIN C, GONG C, LI H. Speaker adaptation research of neural network acoustic model in speech recognition[J]. Computer Applications and Software, 2018, 35(2): 200-205.
金超, 龚铖, 李辉. 语音识别中神经网络声学模型的说话人自适应研究[J]. 计算机应用与软件, 2018, 35(2): 200-205.
[28] HUANG H G, SIM K C. An investigation of augmenting speaker representations to improve speaker normalisation for DNN-based speech recognition[C]//Proceedings of the 2015 IEEE International Conference on Acoustics, Speech and Signal Processing, South Brisbane, Apr 19-24, 2015. Piscataway: IEEE, 2015: 4610-4613.
[29] TAN T, QIAN Y M, YU D, et al. Speaker-aware training of LSTM-RNNs for acoustic modelling[C]//Proceedings of the 2016 IEEE International Conference on Acoustics, Speech and Signal Processing, Shanghai, Mar 20-25, 2016. Piscata-way: IEEE, 2016: 5280-5284.
[30] ROWNICKA J, BELL P, RENALS S. Embeddings for DNN speaker adaptive training[C]//Proceedings of the 2019 IEEE Automatic Speech Recognition and Understanding Workshop, Singapore, Dec 14-18, 2019. Piscataway: IEEE, 2019: 479-486.
[31] VARIANI E, LEI X, MCDERMOTT E, et al. Deep neural networks for small footprint text-dependent speaker verifica-tion[C]//Proceedings of the 2014 IEEE International Con-ference on Acoustics, Speech and Signal Processing, Florence, May 4-9, 2014. Piscataway: IEEE, 2014: 4052-4056.
[32] SNYDER D, GARCIA-ROMERO D, SELL G, et al. X-Vectors: robust DNN embeddings for speaker recognition[C]//Proceedings of the 2018 IEEE International Conference on Acoustics, Speech and Signal Processing, Calgary, Apr 15-20, 2018. Piscataway: IEEE, 2018: 5329-5333.
[33] KHOKHLOV Y Y, ZATVORNITSKIY A, MEDENNIKOV I, et al. R-Vectors: new technique for adaptation to room acoustics[C]//Proceedings of the 20th Annual Conference of the International Speech Communication Association, Graz, Sep 15-19, 2019: 1243-1247.
[34] SHI Y P, HUANG Q, HAIN T. H-Vectors: utterance-level speaker embedding using a hierarchical attention model[C]//Proceedings of the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing, Barcelona, May 4-8, 2020. Piscataway: IEEE, 2020: 7579-7583.
[35] VESELY K, WATANABE S, ZMOLíKOVá K, et al. Se-quence summarizing neural network for speaker adaptation[C]//Proceedings of the 2016 IEEE International Conference on Acoustics, Speech and Signal Processing, Shanghai, Mar 20-25, 2016. Piscataway: IEEE, 2016: 5315-5319.
[36] SARI L, THOMAS S, HASEGAWA-JOHNSON M A. Lear-ning speaker aware offsets for speaker adaptation of neural networks[C]//Proceedings of the 20th Annual Conference of the International Speech Communication Association, Graz, Sep 15-19, 2019: 769-773.
[37] XIE X R, LIU X Y, TAN L, et al. Fast DNN acoustic model speaker adaptation by learning hidden unit contribution features[C]//Proceedings of the 20th Annual Conference of the International Speech Communication Association, Graz, Sep 15-19, 2019: 759-763.
[38] PAN J, WAN G S, DU J, et al. Online speaker adaptation using memory-aware networks for speech recognition[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2020, 28: 1025-1037.
[39] SARI L, MORITZ N, HORI T, et al. Unsupervised speaker adaptation using attention-based speaker memory for end-to-end ASR[C]//Proceedings of the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing, Barcelona, May 4-8, 2020. Piscataway: IEEE, 2020: 7384-7388.
[40] ABDEL-HAMID O, JIANG H. Fast speaker adaptation of hybrid NN/HMM model for speech recognition based on discriminative learning of speaker code[C]//Proceedings of the 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, Vancouver, May 26-31, 2013. Pis-cataway: IEEE, 2013: 7942-7946.
[41] GU D, LI H. Research on speaker adaptation method based on deep neural network[J]. Information Technology and Network Security, 2018, 37(4): 60-64.
古典, 李辉. 基于深度神经网络的说话人自适应方法研究[J]. 信息技术与网络安全, 2018, 37(4): 60-64.
[42] XUE S F, ABDEL-HAMID O, HUI J, et al. Direct adaptation of hybrid DNN/HMM model for fast speaker adaptation in LVCSR based on speaker code[C]//Proceedings of the 2014 IEEE International Conference on Acoustics, Speech and Signal Processing, Florence, May 4-9, 2014. Piscataway: IEEE, 2014: 6339-6343.
[43] MüLLER M, STüKER S, WAIBEL A. Language adaptive multilingual CTC speech recognition[C]//LNCS 10458: Pro-ceedings of the 19th International Conference on Speech and Computer, Hatfield, Sep 12-16, 2017. Cham: Springer, 2017: 473-482.
[44] STADERMANN J, RIGOLL G. Two-stage speaker adaptation of hybrid tied-posterior acoustic models[C]//Proceedings of the 2005 IEEE International Conference on Acoustics, Speech, and Signal Processing, Philadelphia, Mar 18-23, 2005. Pis-cataway: IEEE, 2005: 977-980.
[45] SINISCALCHI S M, LI J Y, LEE C H. Hermitian-based hidden activation functions for adaptation of hybrid HMM/ANN models[C]//Proceedings of the 13th Annual Conference of the International Speech Communication Association, Portland, Sep 9-13, 2012: 2590-2593.
[46] WANG Z Q, WANG D L. Unsupervised speaker adaptation of batch normalized acoustic models for robust ASR[C]//Proceedings of the 2017 IEEE International Conference on Acoustics, Speech and Signal Processing, New Orleans, Mar 5-9, 2017. Piscataway: IEEE, 2017: 4890-4894.
[47] MANA F, WENINGER F, GEMELLO R, et al. Online batch normalization adaptation for automatic speech recognition[C]//Proceedings of the 2019 IEEE Automatic Speech Re-cognition and Understanding Workshop, Singapore, Dec 14-18, 2019. Piscataway: IEEE, 2019: 875-880.
[48] LIU C J, WANG Y Q, KUMAR K, et al. Investigations on speaker adaptation of LSTM RNN models for speech reco-gnition[C]//Proceedings of the 2016 IEEE International Con-ference on Acoustics, Speech and Signal Processing, Shanghai, Mar 20-25, 2016. Piscataway: IEEE, 2016: 5020-5024.
[49] SAMARAKOON L, SIM K C. Factorized hidden layer adaptation for deep neural network based acoustic modeling[J]. IEEE Transactions on Audio, Speech and Language Pro-cessing, 2016, 24(12): 2241-2250.
[50] JIAN X, LI J Y, GONG Y F. Restructuring of deep neural network acoustic models with singular value decomposition[C]//Proceedings of the 14th Annual Conference of the Inter-national Speech Communication Association, Lyon, Aug 25-29, 2013: 2365-2369.
[51] JIAN X, LI J Y, DONG Y, et al. Singular value decompo-sition based low-footprint speaker adaptation and persona-lization for deep neural network[C]//Proceedings of the 2014 IEEE International Conference on Acoustics, Speech and Signal Processing, Florence, May 4-9, 2014. Piscataway: IEEE, 2014: 6359-6363.
[52] SWIETOJANSKI P, RENALS S. Learning hidden unit con-tributions for unsupervised speaker adaptation of neural net-work acoustic models[C]//Proceedings of the 2014 IEEE Spoken Language Technology Workshop, South Lake, Dec 7-10, 2014. Piscataway: IEEE, 2014: 171-176.
[53] XIE X R, LIU X Y, TAN L, et al. BLHUC: Bayesian learning of hidden unit contributions for deep neural network speaker adaptation[C]//Proceedings of the 2019 IEEE International Conference on Acoustics, Speech and Signal Processing, Brighton, May 12-17, 2019. Piscataway: IEEE, 2019: 5711-5715.
[54] DONG Y, YAO K S, HANG S, et al. KL-divergence regula-rized deep neural network adaptation for improved large vocabulary speech recognition[C]//Proceedings of the 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, Vancouver, May 26-31, 2013. Piscataway: IEEE, 2013: 7893-7897.
[55] LIAO H. Speaker adaptation of context dependent deep neural networks[C]//Proceedings of the 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, Van-couver, May 26-31, 2013. Piscataway: IEEE, 2013: 7947-7951.
[56] XIAO L, BILMES J A. Regularized adaptation of discrimina-tive classifiers[C]//Proceedings of the 2006 IEEE International Conference on Acoustics Speech and Signal Processing, Toulouse, May 14-19, 2006. Piscataway: IEEE, 2006: 237-240.
[57] PRICE R, ISO K I, SHINODA K. Speaker adaptation of deep neural networks using a hierarchy of output layers[C]//Proceedings of the 2014 IEEE Spoken Language Technology Workshop, South Lake Tahoe, Dec 7-10, 2014. Piscataway: IEEE, 2014: 153-158.
[58] SWIETOJANSKI P, BELL P, RENALS S. Structured output layer with auxiliary targets for context-dependent acoustic modelling[C]//Proceedings of the 16th Annual Conference of the International Speech Communication Association, Dre-sden, Sep 6-10, 2015: 3605-3609.
[59] MENG Z, LI J Y, GONG Y F. Adversarial speaker adapta-tion[C]//Proceedings of the 2019 IEEE International Confer-ence on Acoustics, Speech and Signal Processing, Brighton, May 12-17, 2019. Piscataway: IEEE, 2019: 5721-5725.
[60] HUANG Z, LI J Y, SINISCALCHI S M, et al. Rapid adapta-tion for deep neural networks through multi-task learning[C]//Proceedings of the 16th Annual Conference of the Inter-national Speech Communication Association, Dresden, Sep 6-10, 2015: 3625-3629.
[61] TóTH L, GOSZTOLYA G. Adaptation of DNN acoustic models using KL-divergence regularization and multi-task training[C]//LNCS 9811: Proceedings of the 18th Interna-tional Conference on Speech and Computer, Budapest, Aug 23-27, 2016. Cham: Springer, 2016: 108-115.
[62] WENINGER F, ANDRéS-FERRER J, LI X W, et al. Listen, attend, spell and adapt: speaker adapted sequence-to-sequence ASR[C]//Proceedings of the 20th Annual Confer-ence of the International Speech Communication Associa-tion, Graz, Sep 15-19, 2019: 3805-3809.
[63] LI K, LI J Y, ZHAO Y, et al. Speaker adaptation for end-to-end CTC models[C]//Proceedings of the 2018 IEEE Spoken Language Technology Workshop, Athens, Dec 18-21, 2018. Piscataway: IEEE, 2018: 542-549.
[64] MENG Z, GAUR Y, LI J Y, et al. Speaker adaptation for attention-based end-to-end speech recognition[C]//Procee-dings of the 20th Annual Conference of the International Speech Communication Association, Graz, Sep 15-19, 2019: 241-245.