A Modified Cross Correlation Double Talk Detector using Variable Threshold for Acoustic Echo Cancellation

Abstract

The presence of echo in the communication systems reduces the speech quality and can be overcome by using Acoustic Echo Cancellers (AEC). Acoustic Echo Canceller is a special device which estimates the echo and subtracts it from the microphone signal. Adaptive filters are used for this purpose. Adaptive filters diverge when in addition to far end signal, near end signal is also present. This situation is known as double talk which is handled by double talk detectors (DTD). Cross correlation based DTD is an attractive approach in which cross correlation between two signals give decision parameter. This parameter is compared to constant threshold to give decision. The constant threshold based double talk detector increases the probability of missed detection and increases the residual error of echo canceller. Thus degrades the performance of Acoustic echo canceller. In this paper, a new variable threshold based double talk detector is proposed in which a threshold value is evaluated with respect to the power of near end signal and far end signal. The proposed method shows the reduced probability of missed detections and mean square error, hence improving AEC performance. For updating of the filter coefficients, Variable Step Size Least Mean Square algorithm is used.