An improved noise reduction algorithm for speech signals using a microphone array

摘要

This paper proposes an improved multichannel Wiener filter (IMWF) algorithm for noise reduction of the speech signals by introducing a new factor when deriving the filter coefficients. This factor, which depends on the multichannel speech presence probability (MC-SPP), helps significantly reduce background noise while suffers only a little speech distortion. In addition, we also propose to modify the minima controlled recursive averaging (MCRA) algorithm, by exploiting shorter time frames and slower statistic change in speech signals, which helps increase the computing speed with almost no degradation in the estimation quality of noise statistics. The performance of our algorithms are simulated and verified for various noise scenarios.