Efficient low delay filtering for residual echo suppression

In telecommunications terminals speech quality is often degraded by acoustic echo. Various approaches to echo cancellation have been proposed and generally involve two separate stages, namely those of adaptive echo cancellation and, as is the focus here, residual echo suppression. Whilst computationally efficient, residual echo suppression approaches based on gain loss control have poor double talk performance. Sub-band approaches give better performance but generally introduce a significant signal delay. This paper reports new experimental work which assesses the performance of three low delay approaches to suppress residual echo through time domain convolution. Results show that a new approach, based on the inverse discrete Fourier transform, performs as well as the existing approaches for both linear and non-linear echo whilst maintaining computational efficiency and low signal delay.