Acoustic echo cancellation in discrete Fourier transform domain based on adaptive combination of adaptive filters

Abstract

Acoustic echo cancellers (AECs) are vital to many of communication systems, including hands-free telephone and videoconference, among others. Recently, adaptive combination of adaptive filters has been presented as an easy but effective method to improve its performance, alleviating different compromises inherent to adaptive filters (responsible to identify the room impulse response). The most important tradeoff is related with the selection of the step size, which involves speed of convergence and residual echo. However, AECs require long adaptive filters, forcing to employ frequency-domain techniques to reduce the computational load and to accelerate the algorithms convergence when colored inputs such speech are presents. In this paper, we present an AEC based on combination of filters in discrete Fourier transform domain. Considering that both the input signal and the cancellation scenario make the behavior of adaptive filters is frequency dependent, our proposal exploits the combination capabilities employing different mixing parameters to separately combine independent spectral regions of two frequency-domain adaptive filters with different step sizes. In this way, our scheme outperforms recent algorithms where only a single combining parameter mixes the overall outputs of two frequency-domain adaptive filters. These advantages are illustrated by means of realistic experiments. © 2013 Acoustical Society of America.