A fast double-talk detection algorithm based on signal envelopes for implementation of acoustic echo cancellation in embedded systems

Abstract

Acoustic echo cancellers based on adaptive filters rely on the adaptive approximation of the echo path. However, the adaptive filter faces the risk of divergence during Double-Talk periods when both echo and near end signal are present. So, the filter adaptation should be stopped during this period; hence the need for a Double-Talk Detection (DTD) algorithm. One of the crucial prerequisites of real-time echo cancellation is that the DTD algorithm must be fast enough to keep up with the adaptive algorithm. The computational complexity of the algorithm is also an important factor when it is implemented in an embedded system. In this paper we present a computationally efficient DTD algorithm which analyses signal envelopes and a few other easily computable values to detect Double-Talk with satisfactory precision and speed. The performance of the algorithm on a realistic Embedded AEC System is also evaluated.