Autocorrelation Based Double Talk Detection System with an NLMS Aczostic Echo Canceler

Abstract

This paper presents an autocorrelation-based double talk detection (DTD) system with a modified normalized least mean squares (NLMS) adaptive filter algorithm for acoustic echo cancellation (AEC). Autocorrelation value of the residual error derived from the difference between the microphone signal and the echo estimation determines the existence of double talk in the speech signal. The algorithms for DTD and adaptive filter for AEC have been tested in terms of probability of missing detection, fraction of times of missing near-end signal and probability of false detection to verify the convergence of the echo canceler in the presence of a near-end signal. The proposed system has been simulated on MATLAB and an application code has been compiled for Xtensa DSP processor. The test results corroborate the reliability of the developed algorithms.