Dual microphone speech dereverberation

This paper addresses the problem of intelligibility and quality improvement for the speech corrupted due to reverberation. Several speech de-reverberation algorithms (both single- and two-channel methods) are studied and compared. Based on this study, we propose some hybrid methods to improve both intelligibility and quality of reverberated speech. The performance of the techniques are examined and compared through both diagnostic rhyme test (DRT) and mean opinion score (MOS). Experimental results with practical data from reverberant situations are used to show the performance of the algorithms. It will be shown that by combining state-of-art correlation shaping, delay-and-sum beamforming, and phase-error-based algorithms, we will reach a hybrid de-reverberation method offers significant improvement in output intelligibility and quality.