A New Fast Double-Talk Detector Based on the Error Variance for Acoustic Echo Cancellation

In order to improve the speech quality in communication systems, acoustic echo cancellation techniques are commonly used to mitigate the deleterious effect of acoustic feedback. In fact, double-talk situations hinder the performance of acoustic echo cancellation when the two speakers in the two ends talk simultaneously. For this reason, double-talk detection is included to control the echo canceler system. In this paper we proposed a new method of double-talk detection based on the error signal variance. Opposed to the previous works where the most of the existing methods are based on a comparison between the received far -end and the microphone observation signals, we accurately account for the variation of the error signal. To evaluate the proposed method, we used acoustic echo cancellation based on the normalized least mean square algorithm. Simulation results indicate the good performance of the proposed double-talk detector.