The speech recognition system for all the Chinese syllables using hidden Markov model

A speech recognition system for all the Chinese syllables is described. The system is a hidden Markov model (HMM)-based recognizer using the initial consonant and the final vowel as the recognition unit with various features derived from linear predictive coding cepstral coefficients. In order to deal with the difficulties introduced by variabilities of speech, the authors transformed a cepstral for a vowel and a multimodel for a consonant. Each element is represented by a hidden Markov model. It is shown that the HMM alone is inadequate in such a difficult task. A syllable recognition accuracy of 93% for a speaker-dependent test is reported. The feasibility of the system is shown