Double talk detection using a psychoacoustic auditory model

Abstract

Successful adaptive echo cancellation in telecommunications depends on a control device called a double talk detector. Double talk refers to the situation that signals at both ends of a communication link are simultaneously active. In the presence of a double talk situation, the role of a double talk detector is to assure convergence and to prevent divergence of the adaptive filter in an echo cancellation system. Following a thorough study of the subject matter, this thesis presents a double talk detection algorithm for a single-channel echo cancellation application using a psychoacoustic auditory model. The model exploits the frequency masking properties of the human auditory system. It performs an analysis of the far-end signal to compute a perceptual threshold, and then removes spectral components below the perceptual threshold to create spectral holes without affecting the perceptual quality of the signal. Double talk conditions are detected by monitoring the spectral levels at the created holes in the near-end input signal. To evaluate the proposed algorithm, results of simulations with real speeches and performance comparisons with certain other double talk detection algorithms are presented. It is shown with simulation results and analysis that the proposed algorithm outperforms those algorithms in overall.