Speech dereverberation based on a recorded handclap

Abstract

A semi-blind framework for the suppression of late speech reverberation is presented. The method is based on spectral subtraction and utilizes a simple recorded handclap to estimate the Power Spectral Density (PSD) of late reverberation. A statistical analysis of measured Room Impulse Responses (RIRs) and recorded handclaps demonstrates the sufficiency of the above estimation. Dereverberation results show that the proposed technique achieves significant reverberation suppression at multiple speaker positions in each room without compromising the quality of the estimated signals.