收稿日期: 2013-06-14
修回日期: 2013-11-04
网络出版日期: 2014-09-15
基金资助
国家自然科学基金（61005019，61273268，90920302）和北京市自然科学基金（KZ201110005005）资助
Data selection method in speaker recognition based on classification of phonemes
Received date: 2013-06-14
Revised date: 2013-11-04
Online published: 2014-09-15
在说话人识别中，有效语音数据的选择是一个重要的预处理环节.常用的数据选择方法根据能量信息的强弱对有效数据进行提取，但在实际情况中能量的高低与语音数据并无必然联系.本文在对传统方法进行分析比较的同时引入语言学知识，提出基于辅音信息的有效数据选择方法.该方法通过对活动语音检测结果中音素识别结果进行分析，保留所有元音，对辅音进行筛选，去除无益于说话人识别的干扰辅音音素，从而实现对有效语 音数据的选取.实验表明，应用该方法得到的说话人识别结果，明显优于传统的基于能量的数据选择算法，如基于G.723.1标准的活动语音检测算法和近期提出的基于交叉熵顺序统计滤波的端点检测算法.
吴蔚澜 , 张卫强 , 刘巍巍 , 田垚 , 陈振锋 , 刘加 , 夏善红 . 说话人识别中基于音素分类的数据选择方法[J]. 中国科学院大学学报, 2014 , 31(5) : 714 -719 . DOI: 10.7523/j.issn.2095-6134.2014.05.019
In speaker recognition, the selection of useful information is an important pre-processing step. Usual ways for selection of the useful information are based on energy. However, between useful information and energy there are no necessary connections. After analying the traditional selection ways, we propose a phoneme decoder based data selection algorithm. Through analysis of the phoneme recognition results, all vowels are kept and some useless consonants are filtered. The speaker recognition experiment results show that the proposed method is superior to the traditional energy-based data selection algorithms such as G.723.1 algorithm and the recently proposed cross entropy based order statistics filtering algorithm.
Key words： speaker recognition; useful information; phoneme decoder; consonant
[1] Kinnunen T, Li H Z. An overview if text-independent speaker recognition: From features to supervectors[J]. Speech Comm, 2010(52):12-40.
[2] Yang X L, Tan B H, Ding J H, et al. Comparative study on voice activity detection algorithm[C]//International Conference on Electrical and Control Engineering. 2010: 599-602.
[3] G.723.1, Annex A:Silence compression scheme[S]. ITU-T, Nov 1996.
[4] Qian Y M, Liu J. Cross-entropy OSF-based voice activity detection algorithm[J]. J Tsinghua Uni, 2009, 49(10): 87-90(in Chinese). 钱彦旻, 刘加. 基于交叉熵顺序统计滤波的语音端点检测算法[J]. 清华大学学报, 2009, 49(10):87-90.
[5] Ramírez J, Segura J C, Benítez C. An effective subband OSF-based VAD with noise reduction for robust speech recognition[J]. IEEE Transaction on Speech and Audio Processing, 2005, 13(6): 1119-1129.
[6] Ramírez J, Segura J C, Benítez C. A new Kullback-Leibler VAD for s peech recognition in noise[J]. IEEE Signal Processing Letters, 2004, 11(2): 266-269.
[7] Juang B H, Rabiner L R. Hidden markov models for speech recognition[J]. Technometrics, 1991, 33(3):251-272.
[8] BenZeghiba M F, Gauvain J L, Lamel L. Context-dependent phone models and models adaptation for phonotactic language recognition[C]//Proceedings of Interspeech 2008.Brisbane, 2008:313-316.
[9] Jesus Antonio Villalba Lopez. Segmentation Experiments for NIST SRE[R]. Brno University of Technology, 2009.
[10] Reynolds D A, Quatieri T F, Dunn R B. Speaker verification using adapted Gaussian mixture models[J]. Digital Signal Processing, 2000, 10: 19-41.
[11] NIST. 2008 NIST Speaker Recognition Evaluation[EB/OL]. (2008-10-04)[2013-05-10]. http://www.itl.nist.gov/iad/mig/tests/spk/2008/index.html.
[12] NIST. 2010 NIST Speaker Recognition Evaluation[EB/OL]. (2010-04-21)[2013-05-10]. http://www.itl.nist.gov/iad/mig/tests/spk/2010/index.html.
[13] 丁爱明.作为说话人识别特征参量的MFCC的提取过程[J].电子工程师, 2006, 32(1):51-53.
[14] Hermansky H.Perceptual linear predictive (PLP) analysis of speech[J]. Journal Acoustical Society of America, 1990, 87(4): 1 738-1 752.
[15] Hermansky H, Hanson B A. Perceptually based linear predictive analysis of speech[C]//Proc of the IEEE International Conference on Acoustics, Speech and Signal Processing. Tampa USA, 1985, X: 509-512.
/
| 〈 |
|
〉 |
