Efficient Speech Dereverberation for Binaural Hearing Aids

An efficient two-stage algorithm for binaural cue preserving speech dereverberation by spectral weighting is presented. The first stage reduces late reverberant speech components where the second stage aims at a suppression of non-coherent early and remaining late reverberant speech components. The proposed system is based on the binaural speech dereverberation algorithm of Jeub et al. [3]. In contrast to that contribution, the realization of the needed delayand-sum beamformer and the blind estimation of reverberation time and time-difference-of-arrivalwith low complexity and low delay is tackled. In addition, the problem of a decreased signal energy for the enhanced signal due to spectral weighting is addressed by an efficient approach.