Joint dereverberation and noise reduction based on acoustic multichannel equalization

Abstract

In many speech communication applications, the recorded microphone signals are often corrupted by both reverberation and noise, which can significantly impair speech quality and intelligibility. While acoustic multichannel equalization techniques can achieve a high dereverberation performance, they may lead to amplification of the additive noise, since the equalization filters are typically designed without taking the presence of noise into account. This paper presents a novel approach to joint dereverberation and noise reduction based on acoustic multichannel equalization (DeNoREq). DeNoREq produces a weighted minimum mean-square error estimate of the clean speech signal, where the weighting parameter trades off between dereverberation and noise reduction performance. Furthermore, an automatic procedure for the selection of the weighting parameter is established. Experimental results for perfectly and erroneously estimated room impulse responses illustrate the effectiveness of the proposed technique in achieving a high dereverberation and noise reduction performance.