基于最大似然线性回归的随机段模型说话人自适应研究

J4 ›› 2014, Vol. 36 ›› Issue (08): 1604-1608.

基于最大似然线性回归的随机段模型说话人自适应研究

晁浩1,2，杨占磊2，刘文举2

(1.河南理工大学计算机科学与技术学院，河南焦作 454000;
2.中国科学院自动化研究所模式识别国家重点实验室，北京100190)

收稿日期:2012-12-19 修回日期:2013-04-03 出版日期:2014-08-25 发布日期:2014-08-25
基金资助:
国家自然科学基金资助项目(91120303，90820303，90820011);国家973计划资助项目(2004CB318105);国家863计划资助项目(20060101Z4073，2006AA01Z194)

Research of speaker adaptation of stochastic
segment models using maximum likelihood linear regression

CHAO Hao1,2,YANG Zhanlei2,LIU Wenju2

(1.School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 454000；2.National Laboratory of Pattern Recognition,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China)

Received:2012-12-19 Revised:2013-04-03 Online:2014-08-25 Published:2014-08-25

摘要/Abstract

摘要：

提出了一种随机段模型系统的说话人自适应方法。根据随机段模型的模型特性，将最大似然线性回归方法引入到随机段模型系统中。在“863test”测试集上进行的汉语连续语音识别实验显示，在不同的解码速度下，说话人自适应后汉字错误率均有明显的下降。实验结果表明,最大似然线性回归方法在随机段模型系统中同样能取得较好的效果。

关键词: 语音识别, 说话人自适应, 最大似然线性回归, 随机段模型

Abstract:

A speaker adaptation method of Stochastic Segment Model (SSM) is proposed. According to the SSM’s characteristics, the theory of Maximum Likelihood Linear Regression (MLLR) method is introduced into the SSMbased systems. Continuous Chinese speech recognition experiment on "863test" test suite shows that the proposed method makes the error rate of Chinese characters decrease obviously under different decoding speeds. Experiment results indicate that the proposal can also improve the recognition performance on the SSMbased systems.

Key words: speech recognition;speaker adaptation;maximum likelihood linear regression;stochastic segment model

晁浩1,2，杨占磊2，刘文举2 . 基于最大似然线性回归的随机段模型说话人自适应研究[J]. J4, 2014, 36(08): 1604-1608.

CHAO Hao1,2,YANG Zhanlei2,LIU Wenju2. Research of speaker adaptation of stochastic
segment models using maximum likelihood linear regression [J]. J4, 2014, 36(08): 1604-1608.

参考文献［10］

［1］	Li Husheng,Liu Jia,Liu Runsheng.Technology of speaker adaptation in speech recogniton and its development trend［J］. Acta Electronica Sinica,2003,31(1):103108. (in Chinese)
［2］	Kimball O,Ostendorf M,Bechwati I.Context modeling with the stochastic segment model［J］. IEEE Transactions on Signal Processing,1992,40(6):15841587.
［3］	Tang Yun, Liu WenJu, Xu Bo, Mandarin digit string recognition based on segment model using posterior probability decoding［J］. Chinese Journal of Computers, 2006,29(4):635642. (in Chinese)
［4］	Tang Yun, Liu WenJu, Zhang Hua. Onepass coarsetofine segmental speech decoding algorithm［C］∥Proc of ICASSP, 2006:441444.
［5］	Zhang Hua, Liu Wenju, Xu Bo. Research on adaptive step decoding in segmentbased LVCSR［C］∥Proc of IEEE NLPKE’07, 2007:463467.
［6］	Gauvain J L, Lee C H. Maximum a posteriori estimation for multivariate Gaussian observations［J］. IEEE Transactions on Speech and Audio Processing, 1994, 2(2):291298.
［7］	Leggetter C,Woodland P.Maximum likelihood linear regression for speaker adaptation of continuous density hidden markov models［J］. Computer Speech and Language, 1995, 9(2):171185.
［8］	Tang Yun, Rose R C. Rapid speaker adaptation using clustered maximumlikelihood linear basis with sparse training data［J］. IEEE Transactions on Audio, Speech & Language Processing, 2008, 16(3):607616.
［9］	Kuhn R, Junqua J C. Rapid speaker adaptation in eigenvoice space［J］. IEEE Transactions on Speech and Audio Processing, 2000, 8(6):695707.
［10］	Luo Jun, Ou Zhijian, Wang Zuoying. Eigenvoicebased MAP fast adaptation in correlation subspaces［J］. Journal of Tsinghua University (Science and Technology), 2005, 8(6):829832. (in Chinese)
［11］	Li C F, Siu M. Training for polynomial segment model using the expectation maximization algorithm［C］∥Proc of ICASSP, 2004:841844.
［12］	Yang Zhanlei, Liu Wenju. A novel path extension framework using steady segment detection for mandarin speech recognition［C］∥Proc of INTERSPEECH, 2010:226229.
	附中文参考文献：
［1］	李虎生，刘加，刘润生. 语音识别说话人自适应研究现状及发展趋势［J］. 电子学报,2003,31(1):103108.
［3］	唐赟,刘文举,徐波. 基于后验概率解码段模型的汉语语音数字串识别［J］. 计算机学报, 2006, 29(4):635642.
［10］	罗骏, 欧智坚, 王作英. 基于相关子空间本征音分析的MAP快速自适应［J］. 清华大学学报（自然科学版），2005, 44(6):829832.

[1]	张开生, 赵小芬. 复杂环境下基于自适应深度神经网络的鲁棒语音识别[J]. 计算机工程与科学, 2022, 44(06): 1105-1113.
[2]	李冠宇，于洪志，李永宏，马宁. 基于决策树的藏语拉萨话三音子模型[J]. J4, 2013, 35(9): 146-150.
[3]	钟金荣,梁宇,林嘉宇. 一种双通道语音增强方法[J]. J4, 2012, 34(5): 172-177.
[4]	包亚萍,郑〓骏,武晓光. 基于HMM和遗传神经网络的语音识别系统[J]. J4, 2011, 33(4): 139-144.
[5]	马震1,谭业武1,陶立慧2,朱茜1. 基于MVDR和ICA的语音识别方法研究[J]. J4, 2010, 32(8): 158-160.
[6]	曹平陈盼章文彬张潮. 基于脉冲神经网络的语音识别方法的初步探究[J]. J4, 2008, 30(4): 139-141.
[7]	丰洪才卢正鼎. 基于置信度的无监督说话人自适应语音识别[J]. J4, 2005, 27(9): 93-96.
[8]	徐静波冉崇森. 语音线性预测技术新探[J]. J4, 2004, 26(5): 93-95.
[9]	张宗红宋彬. 盲的环境补偿方法[J]. J4, 2000, 22(4): 100-102.
[10]	吴江琴高文. HMM方法在手语识别中的应用[J]. J4, 1999, 21(3): 40-44.