Noise Reduction for Noise Robust Feature Extraction for Distributed Speech Recognition

This paper describes the noise robust feature extraction methods developed by France Telecom and Alcatel for the noise robust front-end standardisation of ETSI Aurora. It is shown that both noise reduction methods give a substantial improvement when compared to a standard MFCC feature extraction algorithm for speech recognition in noisy environments. In addition, blind equalisation and feature vector selection were used for further improvement of recognition performance. Results are discussed for the ETSI Aurora 2 task and the SDC-Italian task as well. It was found that the combination of noise reduction with the proposed methods is capable to achieve around 50% reduction of the error rate. In the context of the open ETSI Aurora standardisation, two proposals were submitted based on these methods, they achieved the best results among all the proposals.