Blind Dereverberation for Hearing Aids with Binaural Link

Abstract:
A new two-stage algorithm for binaural dereverberation is proposed which achieves a joint suppression of early and late reverberant speech. All needed quantities are estimated blindly from the reverberant speech and no information about the acoustical environment such as the reverberation time (RT) is required. The first stage of the algorithm is based on a spectral subtraction rule which depends on the spectral variance of the late reverberant speech. The calculation of the spectral variances of the late reverberant speech requires an estimate of the reverberation time. This is accomplished by an efficient algorithm which is based on a maximum likelihood (ML) estimation. In a second stage, the output is further enhanced by a multi-channel Wiener filter. This is derived by a coherence model which takes the shadowing effects of the head into account. The overall binaural input-output processing does not affect the most important binaural cues, i.e., the interaural time difference (ITD) and interaural level difference (ILD). This is important especially for speech enhancement in hearing aids to preserve the ability for source localization in the azimuth plane. Experiments have shown that the new system achieves a significant reduction of early and late reverberation.