Author(s)

Renjie Wu, Lie Chen, Jucai Lin, Jun Yin

Corresponding Author

Renjie Wu

ABSTRACT

Acoustic echo cancellation has conventionally employed all variants known from deterministic adaptive filter design. The presence of double talk makes adaptive algorithm divergent. Meanwhile, a double talk detector with high accuracy and low complexity cannot be implemented easily. In this paper, we explore interesting connections between blind source separation and acoustic echo cancellation, and develop a framework which blind source separation separates mixed signal in double-talk scenario to avoid algorithm diverge instead of double talk detector. The forward BSS is employed as a preprocessor to separate near-end speech while AEC cancels the residual echo. The simulation results are evaluated with ERLE and its performance shows that the proposed framework is effective in double talk scenario.

KEYWORDS

Acoustic echo cancellation, double-talk, blind source separation