Noise-reduced GMDF for acoustic echo cancellation and speech recognition in mobile environments
Jeffrey Lariviere, Rafik A Goubran · 2002
A noise-reduction (NR) enhancement to the generalized multi-delay frequency domain algorithm, suitable for automatic speech recognition (ASR) and acoustic echo cancellation (AEC) is presented. Findings are presented on the ability of the algorithm to suppress car noise from the microphone signal in hands-free mobile telephony in terms of an improvement in echo return loss enhancement (ERLE) and error rates of a digit recognizer from a commercial ASR engine. Minimal computations are involved in the enhancement, e.g. for an additional 11 N/P real multiplies and 8 N/P real additions per output sample, where N is the block size and P is the number of output samples per GMDF iteration, results show an average increase in ERLE of 9dB and an improvement of 15% in digit error recognition rates at low SNR (-5dB).