Technical Report CAIP TR SPEAKER AND ENVIRONMENT ADAPTATION IN CONTINUOUS SPEECH RECOGNITION

Hidden Markov Models HMMs have been used with consider able success in continuous speech recognition It is well known that high accuracy can be obtained when the HMM system is trained and tested in a quiet environment and the speech signal is acquired from a close talking microphone However mismatches between training and testing environment severely degrade erformance Two major sources of mismatches are speaker and environment variability Speaker variation is typ ically caused by di erent speaking styles and other physiolog ical di erences between speakers such as vocal tract lengths etc Environment variability includes channel distortion such as that which a ects telephone speech additive noise and re verberation which results when the microphone is far away from the speaker The goal of this report is to explore di erent adaptation al gorithms that mitigate the e ects of speaker and environmen tal variability for speech recognition The adaptation algorithm closely examined in this report is a Linguistic Tree based Maxi mum Likelihood Linear Regression LT MLLR Speech Recog nition experiments using the LT MLLR for speaker and envi ronment adaptation are given It is shown that the LT MLLR algorithm is superior to other adaptation algorithms discussed For speaker adaptation a reduction is achieved over the baseline word error rate WER using this algorithm In addition it is shown that the use of Matched Filter Array Processing MFA with LT MLLR reduces the WER of distant talking speech with high reverberation In the case when the reverberation time is as high as s the WER is reduced from to a reduction of This work was supported by DARPA Contract DABT C