基于双高斯GMM的特征参数规整和其在语音识别中的应用.pdfVIP

32 4 Vol. 32, No. 4 2006 7 ACTA AUTOMATICA SINICA July, 2006 GMM Æ 1) ( 230027) (E-mail: lrdai@ustc.edu.cn)  (Gaussian mixture model, GMM)  (CDF), CDF ℄℄ Aurora 2 ℄Aurora 3  (Cepstral mean normalization, CMN) ℄ (Cepstral mean and variance normalization, CMVN) Æ—  ℄  TN912 Double Gaussian GMM Based Feature Normalization and Its Application in Speech Recognition LIU Bo DAI Li-Rong WANG Ren-Hua DU Jun LI Jin-Yu (Department of Electronic Engineering Information Science, University of Science Technology of China, Hefei 230027) (E-mail: lrdai@ustc.edu.cn) Abstract In this paper, a new feature normalization approach based on double Gaussian mixture model is proposed. Since speech features in noisy environments usually follow bi- modal distributions, to fully utilize this characteristic we represent the cumulative density function (CDF) of the features with a more delicate Gaussian mixture model. Finally, fea- ture normalization process is performed according to the estimated CDF to improve speech recognition performance. Experimental results on Aurora 2 and Aurora 3 tasks show that the performance of our method is much better than those of the conventional cepstral mean nor- malization (CMN) and cepstral mean and variance normalization (CMVN) methods, and is comparable to that of the histogram equalization method, which is a non-parametric method. Key words Speech recognition, front-end, noise-robustness, speech feature normalization, histogram equalization 1) Æ , Æ (863 )(2004A