Speech dereverberation in multisource environment using LCMV filter

In this paper, a novel multi channel speech dereverberation method in a multisource environment is proposed. The primary contribution of this work is the utilization of a multi channel linearly constrained minimum variance (LCMV) filter for speech dereverberation in the presence of multiple sources. An initial estimate of the desired speaker signal is obtained using a sparse matrix structure obtained herein. A multi channel LCMV filter is then used to obtain the final estimates of the desired signal by applying constraints on both early and late part of individual frames of the short term speech signal. Experiments on speech dereverberation and distant speech recognition are conducted at various direct to reverberant ratios to evaluate the performance of the proposed method. Performance improvements are noticable when compared to other multi channel speech dereverberation methods in the literature.