Parallel neural networks for speaker-independent all-Chinese-syllable speech recognition

The paper presents a parallel neural networks approach to speaker-independent all-Chinese-syllable speech recognition. The author uses neural networks to recognize 59 speech units, 22 initial consonants, 36 syllable finals and a background noise, for pattern division. Each binary classifier recognizes a speech unit to discriminate P_i from ~P_i (NoT P_i). The utterances from 137 male speakers are used to train and the utterances from other 11 male speakers are used to recognize. It achieved a recognition correct rate of 66.14% for Chinese syllables, 73.06% for initial consonants and 84.6% for syllable finals. They are rather good without grammer