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

Equalization of room impulse responses is an straightforward approach for dereverberation of speech signals in a hands-free scenario. The wellknown least-squares equalization filter minimizes the euclidian distance between the concatenated system of room impulse response (RIR) and equalizer and a given target system. This target system usually is chosen as a delayed discrete pulse, a delayed band-pass or a delayed high-pass filter. In this contribution we address the choice of the this delay which has to be introduced w.r.t. maximum dereverberation performance. Since designing one equalizer (EQ) for each possible delay and choosing the best one is computationally inefficient we investigate the dependence of the optimum EQ delay on various parameters describing RIRs. A high correlation was found between the so-called central time of the room impulse response and the optimum EQ delay. Since the central time can be determined based on estimates of the initial peak of the RIR and the room reverberation time, we propose to use a very short filter for system identification and an estimate of the room reverberation time to identify the optimum equalizer delay.