Effect of Time Duration and Intrinsic Features for English Phoneme Recognition

Abstract

This paper describes methods to improve the performance of English phoneme recognition from linguistic view points. The methods include exploiting time duration information in hidden Markov model (HMM), intrinsic feature space for vowel. The time duration constraint imposed on states of the phoneme HMM can improve its recognition rate significantly for phoneme data in continuous pseech. As intrinsic feature spaces for vowel, formants and the time derivative are employed. They improve the phoneme recognition rate considerably compared with the commonly used LPC cepstral coefficients.