No abstract

It is well known that noise reduction schemes are beneficial in ASR to reduce training-test mismatch due to noise. However, a significant mismatch may still remain after noise reduction, especially in the non-speech portions of the signals. To reduce the impact of this mismatch, two methods for discarding non speech acoustic vectors at recognition time are investigated: variable frame rate processing and voice activity detection. Experiments are discussed for Aurora 2 and for SpeechDat Car Italian. Results show that both methods are highly effec tive for SpeechDat Car Italian. However, for Aurora 2, feature vector selection based on voice activity detection hardly gives a benefit, while variable frame rate processing actually lowers recognition accuracy somewhat. Several possible explanations o f the different results observed for the two databases are dis cussed

This paper describes the noise robust feature extraction meth ods 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 im provement when compared to a standard MFCC feature ex traction algorithm for speech recognition in noisy environ ments. In addition, blind equalisation and feature vector se lection 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 combi nation of noise reduction with the proposed methods is capa ble to achieve around 50% reduction of the error rate. In the context of the open ETSI Aurora standardisation, two propos als were submitted based on these methods, they achieved the best results among all the proposals.