Continuous Speech Recognition with Phase-corrected Rasta

Phase-corrected RASTA is a new technique for channel normalisation that consists of classical RASTA filtering followed by a phase correction operation. In this manner, the channel bias is as effectively removed as with classical RASTA, without introducing a left-context dependency. The performance of the phase-corrected RASTA channel normalization technique was evaluated for a continuous speech recognition task. Using context-independenthidden Markov models we found that phase-corrected RASTA reduces the bestsentence word error rate (WER) by 10% compared to classical RASTA. Using a version of phase-corrected RASTA suited for real-time implementation WER was reduced by 7:5% compared to classical RASTA.