Joint Dereverberation and Separation of Reverberant Speech Mixtures

Abstract

This paper proposes a joint dereverberation and separation method for speech mixtures. In the literature, several studies consider the problems of blind separation and blind dereverberation as two separate problems and propose a solution to each problem. The dereverberation method consists to estimate the log of maximum likelihood parameter which depends on three parameters. So the reverberation algorithm returns to jointly maximize the likelihood function. The bivariate empirical mode decomposition (BEMD) algorithm combined with complex independent component analysis by entropy bound minimization (ICA-EBM) technique is proposed as an alternative to separate convolutive mixtures of speech signals reverberant environnement. The proposed algorithm is considered as a connection in tandem of the dereverberation network and separation network to estimate the source signals. The performance of the proposed approach is tested on real speech sounds chosen from available databases. Two sets of room impulse responses with reverberation times of 0.3 and 0.5 s, with a total of 410 test samples for each reverberation condition are simulated. The proposed method is compared to the conditional separation and dereverberation (CSD) method and Frequency domain blind source separation (FDBSS) method using objective measures which are segmental signal-to-reverberation ratio (SRRseg), signal-to-interference ratio (SIR) and direct-to-reverberationr (DRR).