A double-talk detection technique using the correlation behaviour of the gradient vector for acoustic echo cancellation

In this paper, an improved adaptive filtering algorithm is introduced for acoustic echo cancellation (AEC) in the double-talk (DT) situation, based on a new DT detector (DTD). The proposed DTD aims to distinguish between the DT situation and the abrupt change in the acoustic echo path (ACAEP) by employing the correlation behaviour of the gradient vector. In addition, another DTD is used to discriminate the DT situation from the steady-state (SS). The adaptation gain of the normalised least mean square (NLMS) algorithm is controlled by these proposed detectors so that it results in fast convergent rate when there is an ACAEP and slow convergent rate during the DT situation and the SS period. Simulation results based on white noise sequences support these ideas.