Compensation of Channel Dist Frequency D

This paper addresses the problem of channel effect in the line spectrum frequency (LSF) domain. The channel effect can be expressed in terms of the channel phase. The speech signal is represented by its inverse filter derived from LP analysis. Then the mean normalization on the inverse filters is introduced for removing the channel distortion. Further study indicates that the mean normalization on the inverse filters becomes the mean subtraction in phase domain. Based on this finding, two methods are proposed to compensate the channel effect. Experiments on simulated channel distorted speech are conducted to evaluate the effectiveness of the proposed methods. The experimental results show that the proposed methods can give significant improvements in speech recognition performance. The performance of the proposed methods is comparable to that of CMN in using cepstral coefficients.