Phase sensitive model adaptation for noisy speech recognition

A computer-readable medium comprising instructions that, when executed by a processor, perform the following acts:

• a) receive a distorted speech utterance;
• b) initialize a convolutive distortion mean vector, an additive distortion mean vector, and a diagonal covariance matrix;
• c) update parameters of a Hidden Markov Model used in connection with speech recognition based at least in part upon the convolutive distortion mean vector, the additive distortion mean vector, and the diagonal covariance matrix;
• d) decode the received speech utterance using parameters of the Hidden Markov Model updated in act c);
• e) re-estimate the additive distortion mean vector and the convolutive distortion mean vector;
• f) update the parameters of the Hidden Markov Model based at least in part upon the re-estimated additive distortion mean vector and the re-estimated convolutive distortion mean vector;
• g) determining content of the received speech utterance through use of the Hidden Markov Model and the updated parameters of act f).