Speech recognition with compensation for both convolutive distortion and additive noise

摘要

A method of speech recognition with compensation is provided by modifying HMM models trained on clean speech with cepstral mean normalization. For each speech utterance the MFCC vector is calculated for the clean speech database. This mean MFCC is added to the original models. An estimate of the background noise is determined for a given speech utterance. The model mean vectors adapted to the noise are determined. The mean vector over the noisy speech space is determined and this is removed from the model mean vectors adapted to noise to get the target model.