Multi-Channel Speech Enhancement using a Psychoacoustic Approach for a Post-Filter

Abstract:
Multi-channel systems containing a Delay&Sum-Beamformer have been previously combined with post-filters to achieve higher speech enhancement in noisy conditions. Problems occuring in the lower frequency regions were compensated by applying aWiener weighting rule or spectral subtraction. We extend this by referring to the log-STSA rule of Ephraim-Malah. Furthermore, we introduce a psychoacustically motivated post-filter based on the one presented by Gustafsson for single-channel systems. Relying on Li’s subband approach we derive an algorithm which exploits the advantage of multi-channel systems in order to estimate those quantities necessary for computing the psychoacoustic weighting rule.