Double-talk detection using the singular value decomposition for acoustic echo cancellation

One of the major problems in voiced communication systems is the presence of acoustic echoes generated from the coupling between the loudspeaker and the microphone. In this paper, a new method of Double-Talk Detection (DTD) for Acoustic Echo Cancellation (AEC), based on the Singular Value Decomposition (SVD), is proposed. Usually, the performances of the AEC, which is based on adaptive filtering, degrade seriously in the presence of speech issued from the near-end speaker (double-talk). Then, Double Talk Detection system must be added to AEC, for controlling the adaptation of the adaptive filter coefficients. For this purpose, we introduce the SVD of the far-end signal for detecting the double-talk periods. The obtained results, using TIMIT database, show that the proposed method outperforms the classical Geigel algorithm and Normalized Cross-Correlation (NCC) algorithm.