AN ADAPTIVE FILTERING METHOD FOR SPEECH PARAMETER ENHANCEMENT

This paper considers the estimation of speech parameters and their enh ancement using an approach based on the estimation-maximization (EM) a lgorithm, when only noisy speech data is available. The distribution o f the excitation source for the speech signal is assumed as a mixture of two Gaussian probability distribution functions with differing vari ances. This mixture assumption is experimentally valid for removing th e residual excitation signal. The assumption also is found to be effec tive in enhancing noise-corrupted speech. We adaptively estimate the s peech parameters and analyze the characteristics of its. excitation so urce in a sequential manner. In the maximum likelihood estimation sche me we utilize the EM algorithm, and employ a detection and an estimati on step for the parameters. For speech enhancement we use Kalman filte ring for the parameters obtained from the above estimation procedure. The estimation and maximization procedures are closely coupled. Simula tion results using synthetic and real speech vindicate the improved pe rformance of our algorithm in noisy situations, with an increase of ab out 3 dB in terms of output SNR compared to conventional Gaussian assu mption. The proposed algorithm also may be noteworthy in that it needs no voiced/unvoiced decision logic, due to the use of the residual app roach.