Estimation of the Optimum System Delay for Speech Dereverberation by Inverse Filtering

Equalization of room impulse responses is an attractiveapproach for dereverberation of speech signals in ahands-free scenario. In this contribution we address thechoice of the delay which has to be introduced in least-squares equalization approaches for a maximum amountof dereverberation. Since designing one equalizer (EQ)for each possible delay and choosing the best one iscomputationally inefficient we evaluate the dependenceof the optimum equalizer delay of various measurescharacterizing room impulse responses (RIRs). A highcorrelation was found between the so-called central timeof the room impulse response and the optimum equalizerdelay. Since the central time can be determined basedon estimates of the initial peak of the RIR and theroom reverberation time, we propose to use a veryshort filter for system identification and an estimate ofthe room reverberation time to identify the optimumequalizer delay. The proposed approach prevents a lowperformance of the equalizer which may occur for animproperly chosen delay by automatically estimating theoptimum delay.