Incorporating the human hearing properties into multichannel speech enhancement

In this paper we propose perceptually motivated multichannel system for speech dereverberation and noise suppression. We modify a conventional generalized sidelobe canceler (GSC) structure by introducing perceptual weighting factor into noise cancellation loop. The purpose of this factor is to shape residual noise according to masking effects and to minimize the speech leakage. We implement the GSC structure using a delay-and-sum beamformer and blocking matrix that depends on second order statistics only. Therefore we do not require exact estimates of the channel transfer functions nor its ratios. Finally a comparative evaluation of the proposed methods is performed using objective speech quality measures. Results show that the novel method outperforms conventional one providing better speech quality.