Joint acoustic echo cancellation and transfer function GSC in the frequency domain

In this paper we present two joint acoustic echo and noise cancellation schemes implemented in the frequency domain and used for hands-free communication. In several past contributions equivalent time domain schemes were proposed. However, the frequency domain allows better convergence performance regardless of the condition number of the correlation matrix of the input data, and therefore is more suitable for speech processing. The first joint scheme we propose contains a multi-channel acoustic echo canceller (AEC) followed by a beamformer as a second stage (this scheme is denoted AEC-BF). The second scheme contains a beamformer followed by a single channel AEC as a postfilter (denoted BF-AEC). Both schemes include the recently proposed transfer function generalized sidelobe canceller (TF-GSC) beamformer and a block-LMS AEC. The performance of both schemes is evaluated through a series of simulations, using real speech recordings in both room and car environment, and under different types of noise signals. The experimental results show that the AEC-BF scheme usually outperforms the BF-AEC scheme.