Abstract
We propose a method to recover the original sound waveform from the observed sound waveform without using the inverse filtering of the room transfer function. Reverberation is caused by the echo element of the original sound signal. In this study, short time frame signals subdivided reverberant speech by time windowing, and we attempted to reduce the echo influence by subtracting previous frame signals from the current frame signal. Weighing coefficient and adjusted time for the subtractions are determined by correlation between past and current signals. In the case that reverberation time is about 1 sec, the result of simulation shows good recovery of original speech.
