ISOLATED MANDARIN BASE-SYLLABLE RECOGNITION BASED UPON THE SEGMENTAL PROBABILITY MODEL

In this correspondence, a segmental probability model (SPM) is propose d for fast and accurate recognition of the highly confusing isolated M andarin base-syllables by deleting the state transition probabilities of continuous density hidden Markov models (CHMM), abandoning the dyna mic programming process, letting the states equally segment the base-s yllables deterministically, and using several special approaches to im prove the accuracy and speed. This is achieved by considering the spec ial characteristics of the target vocabulary.