Towards multi-microphone speech dereverberation using spectral enhancement and statistical reverberation models

In speech communication systems the received microphone signals are degraded by room reverberation and ambient noise. Reverberant speech can be separated into two components, viz. an early speech component and a late reverberant speech component. In this paper a multichannel dereverberation algorithm is proposed to suppress late reverberation. Specifically, we employ a minimum variance distortionless beamformer and a single-channel MMSE estimator, which operates on the beamformer's output signal. The so-called late reverberant spectral variance (LRSV) required by the MMSE estimator can be estimated using i) the beamformer's output signal or ii) the received microphone signals. In this contribution we investigate both approaches and show how a priori knowledge of the reverberant sound field can be exploited to improve the LRSV estimation. Advantages and disadvantages of the LRSV estimators are discussed, and experimental results using simulated reverberant speech are presented.