Optimizing Hidden Markov Models for Chinese An-set Syllables

Speech recognition for Chinese relied very much on the recognition of Chinese syllables and there are altogether 1345[7] syllables in it. If we take tones into considerations, the number of syllables can reduce to 408 base syllables, with different tones, in which it can further divided into 38 confused set. Among those sets, the Chinese An-set is considered as one of the major confused syllable set. Thus, the recognition of Chinese An-set syllables is very important to the Chinese recognition. In this paper, we proposed a new training approach based on maximum model distance (MMD) for HMMs to train the Chinese An-set syllables. Both the speaker-dependent and multi-speaker experiments on the confused Chinese An-set showed that significant error reduction can be achieved through the proposed approach.