Double Talk Detectors with non-Stationary and Noisy Signals

Abstract

An acoustic echo cancellation (AEC) application is considered an interesting process in a single-talk condition. However, when the double talk (DT) signal is available simultaneously with this undesired echo problem, the AEC performance is severely deteriorated due to the rapid deviation of the filter coefficients. Consequently, incorporating a double talk detector (DTD) module with AEC systems is necessary to halt and handle the filter adaptation process during double talk detection. Instead, echo-path changes are basically common in AEC systems, which is one of the main design issues of the logical control of DTD methods. The DTD control can be quite complicated in discriminating the near-end speaker from time-variant echo-path change. In this paper, we introduce several DTD methods and we aim to show the effectiveness and the reliability of these algorithms to differentiate between echo-path variations and double-talk situations that are investigated in our numerical results. In this context, we propose a new criterion to detect echo path changes combined with the cheap normalized cross-correlation (CNCC) double talk detector. The obtained simulation and numerical experiments appear the superior performance of our proposed method in front of DT periods and non-stationary (NS) phases.