非线性统计匹配用于子带鲁棒语音识别

孙暐; 吴镇扬; 刘海滨

非线性统计匹配用于子带鲁棒语音识别

计量
- 文章访问数: 2440
- HTML全文浏览量: 86
- PDF下载量: 753
- 被引次数: 0
出版历程
- 收稿日期: 2004-08-05
- 修回日期: 2005-04-21
- 刊出日期: 2006-03-19

Nonlinear Statistical Matching for Subband Robust Speech Recognition

摘要

摘要: 由于语音信号的多变性，识别系统的性能极易受噪声环境的影响而导致性能下降。该文以听觉试验为基础，提出一种新的非线性独立子带隐马尔可夫模型(HMM)最大后验统计匹配算法。该算法依据人耳感知的频选性，根据各子带噪声特点采用统计匹配、MAP估计和HMM/MLP非线性映射来补偿噪声环境的影响。实验表明该算法明显改善了识别系统在噪声环境下的性能。
- 语音识别;隐马尔可夫模型;最大后验估计;听觉场景分析
Abstract: The performance of the speech recognition systems is deteriorated dramatically under noise condition for variation of speech signal. According to the auditory tests, this paper proposes a new nonlinear sub-band Maximum A Posteriori (MAP)statistical matching algorithm based on the independent sub-band analysis. According to the perception of humans ear and noise feature of different frequency-bands, the algorithm compensates the effects of noise with statistical matching, MAP estimation and HMM/MLP nonlinear mapping. The test shows that the proposed algorithm improves the recognition performance notably under noise condition.

HTML全文

参考文献(1)

Cooke M, Morris A, Green P. Missing data techniques for robust speech recognition[C][J].ICASSP97, Munich, Germany.1997, vol 2:863-[2]Diakoloukas V D, Digalakis V V. Maximum-likelihood stochastic-transformation adaptation of hidden Markov models[J].IEEE Trans. on Speech and Audio Processing.1999, 7(2):177-[3]Siohan O, Chesta C, Lee C -H. Hidden Markov model adaptation using maximum a posteriori linear regression[C]. In Workshop on Robust Methods for Speech Recognition in Adverse Conditions, Tampere, Finland, 1999: 147150. .[4]Gales M, Young S. Cepstral parameter compensation for HMM recognition in noise[J]. Computer Speech and Language, 1993, 12(3)：231.239.[5]Sharma S R. Multistream approach to robust speech recognition[D/D]. Oregon Graduate Institute of Science and Technology, 1999.10.[6]Tibrewala S, Hermansky H. Subband based recognition of noisy speech[C][J].ICASSP97, Munich, Germany.1997, vol 2:1255-[7]Ji M, Smith F J. A probabilistic union model for subband based robust speech recognition[C]. ICASSP'00, Istanbul, Turkey, 2000, vol 3: 1787.1790.[8]孙暐, 吴镇扬, 刘海滨等. 并行子带HMM最大后验概率自适应非线性类估计算法[J]. 电路与系统, 录用待刊.[9]Allen J B. How do humans process and recognize speech[J]. IEEE Trans. on Speech and Audio Processing, 1994, 2(4): 567577. .[10]Dempster A P, Laird N M, Rubin D B. Maximum likelihood estimation from incomplete data[J]. J Royal Statistical Society,Serials B, 1977, 39(1): 138. .[11]Ris C, Dupont S. Assessing local noise level estimation methods: application to noise robust ASR[J].Speech Communication.2001, 34(1-2):141-[12]Hirsh H G.. Estimation of noise spectrum and its application to SNR estimation and speech enhancement. Technical ReportTR-93-012, International Computer Science Institute, Berkeley,USA, 1993.[13]Mak B. A mathematical relationship between fullband and multiband mel-frequency cepstral coefficients[J]. IEEE Signal Processing Letters, 2002, 9(8): 241244.

施引文献

资源附件(0)

访问统计