Blind source separation of mixed speech in a high reverberation environment

Blind source separation (BSS) is a technique for estimating and separating individual source signals from a mixed signal using only information observed by each sensor. BSS is still being developed for mixed signals that are affected by reverberation. In this paper, we propose combining the BSS method that considers reverberation proposed by Duong et al. with the BSS method reported by Sawada et al., which does not consider reverberation, for the initial setting of the EM algorithm. This proposed method assumes the underdetermined case. In the experiment, we compare the proposed method with the conventional method reported by Duong et al. and that reported by Sawada et al., and demonstrate the effectiveness of the proposed method.