Abstract

A powerful technique for modelling the temporal structure and variability in speech is one called hidden Markov modelling. This is a probabilistic pattern-matching approach that models a time-sequence of speech patterns as output of a stochastic or random process. In speech processing, a word is analyzed in 10-30ms time frames and for each frame, we extract a set of coefficients which is called linear predictive coefficients (LPC) and convert one to a cepstral vector. Through a vector quantizer, the vector is mapped to a observation and using hidden Markov model (HMM) to estimate and recognise the observation by Baum-Welch algorithms. The results show that we should used the HMM to recognise Vietnamese speech.