[1]吕勇,吴镇扬.基于隐马尔可夫模型与并行模型组合的特征补偿算法[J].东南大学学报(自然科学版),2009,39(5):889-893.[doi:10.3969/j.issn.1001-0505.2009.05.004]
 Lü Yong,Wu Zhenyang.Feature compensation algorithm based on hidden Markov model and parallel model combination[J].Journal of Southeast University (Natural Science Edition),2009,39(5):889-893.[doi:10.3969/j.issn.1001-0505.2009.05.004]
点击复制

基于隐马尔可夫模型与并行模型组合的特征补偿算法()
分享到:

《东南大学学报(自然科学版)》[ISSN:1001-0505/CN:32-1178/N]

卷:
39
期数:
2009年第5期
页码:
889-893
栏目:
信息与通信工程
出版日期:
2009-09-20

文章信息/Info

Title:
Feature compensation algorithm based on hidden Markov model and parallel model combination
作者:
吕勇 吴镇扬
东南大学信息科学与工程学院,南京 210096
Author(s):
Lü Yong Wu Zhenyang
School of Information Science and Engineering, Southeast University, Nanjing 210096, China
关键词:
语音识别 特征补偿 隐马尔可夫模型 并行模型组合
Keywords:
speech recognition feature compensation hidden Markov model parallel model combination
分类号:
TN912.34
DOI:
10.3969/j.issn.1001-0505.2009.05.004
摘要:
提出了一种基于隐马尔可夫模型和并行模型组合的特征补偿算法.首先,利用一个包含较多状态的隐马尔可夫模型来描述全部单词特征向量的分布.然后,根据静音段估计的噪声均值和方差,采用并行模型组合方法调整隐马尔可夫模型的均值向量和协方差矩阵,使之与识别环境相匹配.最后,根据基于状态转移矩阵压缩的前向后向算法计算隐马尔可夫模型的后验概率,并通过最小均方误差准则估计纯净语音特征向量.实验结果表明,该算法能够更加准确地估计纯净语音特征向量,其性能明显优于基于高斯混合模型的特征补偿算法; 状态转移矩阵压缩算法可以在不影响补偿精度的前提下,显著减少前向后向算法的计算量.
Abstract:
A feature compensation algorithm based on hidden Markov model(HMM)and parallel model combination(PMC)is presented. Firstly, a HMM composed of a number of states is employed to represent the distribution of the speech features of all words. Then, according to the mean and covariance of noise from noise-only frames, the mean vectors and covariance matrices of the HMM are transformed to the testing condition by the PMC method. Finally, the posterior probability of HMM is computed by the forward-backward algorithm based on the compression of the state transition matrix, and the clean speech feature is calculated by the minimum mean squared error method. The experimental results show that the proposed algorithm can restore the clean speech feature more accurately and outperforms the feature compensation algorithm based on Gaussian mixture model(GMM). Besides, the state transition matrix compression method can greatly reduce the computational cost of the forward-backward algorithm without decreasing the compensation performance.

参考文献/References:

[1] Nasersharif B,Akbari A.SNR-dependent compression of enhanced Mel sub-band energies for compensation of noise effects on MFCC features [J]. Pattern Recognition Letters,2007,28(11):1320-1326.
[2] 赵蕤,王作英.语音识别中信道和噪音的联合补偿[J].声学学报,2006,31(5):466-470.
  Zhao Rui,Wang Zuoying.Joint compensation of noise and channel in speech recognition [J].Acta Acustica,2006,31(5):466-470.(in Chinese)
[3] Cui X,Alwan A.Noise robust speech recognition using feature compensation based on polynomial regression of utterance SNR [J]. IEEE Transactions on Speech and Audio Processing,2005,13(6):1161-1172.
[4] Barreaud V,Illina I,Fohr D.On-line stochastic matching compensation for non-stationary noise [J]. Computer Speech and Language,2008,22(3):207-229.
[5] Moreno P J.Speech recognition in noisy environments [D].Pittsburgh,Pennsylvania,USA:Carnegie Mellon University,1996:79-126.
[6] Kim W,Kwon O,Ko H.PCMM-based feature compensation schemes using model interpolation and mixture sharing [C] //IEEE International Conference on Acoustics,Speech,and Signal Processing.Montreal,Canada,2004:989-992.
[7] Kim W,Hansen J H L.Feature compensation in the cepstral domain employing model combination [J]. Speech Communication,2009,51(2):83-96.
[8] Sasou A,Asano F,Nakamura S,et al.HMM-based noise-robust feature compensation [J]. Speech Communication,2006,48(9):1100-1111.
[9] Gales M J F,Young S J.Robust speech recognition in additive and convolutional noise using parallel model combination [J].Computer Speech and Language,1995,9(4):289-307.
[10] 孙暐,吴镇扬.基于独立感知理论的鲁棒语音识别算法[J].东南大学学报:自然科学版,2005,35(4):506-509.
  Sun Wei,Wu Zhenyang.Robust speech recognition algorithm based on fletcher-allen principle [J]. Journal of Southeast University:Natural Science Edition,2005, 35(4):506-509.(in Chinese)
[11] Kim D,Yook D.Linear spectral transformation for robust speech recognition using maximum mutual information [J].IEEE Signal Processing Letters,2007, 14(7):496-499.

相似文献/References:

[1]赵力,刘怡龙,邹采荣,等.基于VQ-HMM的无教师说话人自适应方法[J].东南大学学报(自然科学版),2001,31(2):23.[doi:10.3969/j.issn.1001-0505.2001.02.006]
 Zhao Li,Liu Yilong,Zou Cairong,et al.An Unsupervised Speaker Adaptation Method Based on VQ-HMM[J].Journal of Southeast University (Natural Science Edition),2001,31(5):23.[doi:10.3969/j.issn.1001-0505.2001.02.006]
[2]孙暐,吴镇扬.基于独立感知理论的鲁棒语音识别算法[J].东南大学学报(自然科学版),2005,35(4):506.[doi:10.3969/j.issn.1001-0505.2005.04.002]
 Sun Wei,Wu Zhenyang.Robust speech recognition algorithm based on fletcher-allen principle[J].Journal of Southeast University (Natural Science Edition),2005,35(5):506.[doi:10.3969/j.issn.1001-0505.2005.04.002]

备注/Memo

备注/Memo:
作者简介: 吕勇(1979—),男,博士生; 吴镇扬(联系人),男,教授,博士生导师,zhenyang@seu.edu.cn.
基金项目: 国家重大基础研究发展计划(973计划)资助项目(2002CB312102)、国家自然科学基金资助项目(60672094).
引文格式: 吕勇,吴镇扬.基于隐马尔可夫模型与并行模型组合的特征补偿算法[J].东南大学学报:自然科学版,2009,39(5):889-893. [doi:10.3969/j.issn.1001-0505.2009.05.004]
更新日期/Last Update: 2009-09-20