Combined acoustic echo and noise reduction using GSVD-based optimal filtering

This paper describes two schemes for combining acoustic echo and noise reduction using a GSVD-(generalized singular value decomposition-) based optimal filtering technique. The GSVD-based filtering technique is a signal enhancement technique which has previously been proposed for noise reduction in multi-microphone speech signals. In many speech communication applications however also a far-end echo source is present. Therefore a combined echo and noise reduction scheme is needed. The first scheme combines a standard multi-channel adaptive echo canceller with the GSVD-based noise reduction technique. The second scheme incorporates the far-end echo reference directly into the GSVD-based signal enhancement technique without cancelling the echo in every microphone signal. The two different schemes are compared with regard to performance and computational complexity