A Novel Approach for Dereverberation of Speech Signals

Abstract

In this paper, we propose a novel higher-order statistics based approach to estimate the room impulse response (RIR) of a reverberating auditorium. The RIR is then used to dereverberate the speech signal. It is assumed that the RIR does not change for a brief period of time called the coherence time. Once the acoustic environment changes, a set of low power third-order white pilot samples are used to estimate the RIR. The log-magnitude and phase information of the RIR are retrieved in the bispectrum domain of the received pilot signal. The discrete wavelet transform (DWT) based technique is used to denoise the speech signal. After denoising, the effect of reverberation is removed from the corrupted signal using inverse filtering by the estimated RIR.