Improved nonlinear residual echo suppression based on partial echo path modeling DTD

Abstract

Hand-free systems are widely deployed in voice communication where acoustic echo is inevitable. Acoustic echo cancellation (AEC) is effective to mitigate acoustic echo. In some applications, serious nonlinear acoustic echo presents due to nonlinear components in handhold devices. Since nonlinear echo cannot be satisfactorily cancelled with linear AEC, nonlinear residual echo suppression (NRES) module is introduced to suppress nonlinear residual echo. Considering the NRES module will distort near-end speech, a novel doubletalk detection (DTD) algorithm is proposed based on partial channel modeling to detect near-end speech in nonlinear circumstance. In addition, a structure integrating the new DTD and the NRES module is presented to improve the performance of AEC in nonlinear system. Simulations confirm the effectiveness of the new DTD algorithm and the structure integrating the DTD and the NRES.