Stereo Acoustic Echo Cancellation Based on Maximum Likelihood Estimation With Inter-Channel-Correlated Echo Compensation

초록

This paper presents batch and online algorithms of a stereo acoustic echo cancellation (SAEC) method. In SAEC, the non-uniqueness problem causes performance degradation, especially for highly coherent far-end signals. In our method, this problem can be avoided without an additional decorrelation preprocessor or multi-microphones by overestimating far-end echoes and compensating for the overestimated inter-channel-correlated echo to obtain a desired echo-canceled signal. In addition, our method is based on the maximum likelihood estimation (MLE) criterion of the echo-canceled signal under the assumption that the signal in the time-frequency domain follows a zero-mean complex Gaussian distribution with a time-varying variance. Furthermore, a variable forgetting factor based on the cross spectral density (CSD) between the echo-canceled signal and a far-end echo is presented in the online algorithm to improve the convergence of adaptive filters with a high cancellation performance when converged. Experimental results under various conditions demonstrate that the proposed method can successfully perform SAEC even in the presence of inter-channel correlation, double-talk, and abrupt echo path changes.

키워드