A Robust Residual Echo Suppression Algorithm Even During Double Talk

Abstract

Adaptive echo cancellation (AEC) is an essential part in hand-free full-duplex communication to cancel the annoying echo. However, residual echo always remains even after linear acoustic echo canceller (LAEC) for the misalignment between the room impulse response and the adaptive filter coefficients. In practical applications, AEC algorithm followed by residual echo suppression (RES) is a general strategy. This paper proposes a new algorithm to estimate the residual echo power spectral density after LAEC based on the statistic normalized correlation between the output after LAEC and the estimated echo. The experimental results verify the good performance of the algorithm in terms of return loss enhancement (ERLE) and speech-to-speech-distortion power ratio(SSDR).