The study of recognition of isolate monosyllable Chinese speech using VQ and HMM

A way of recognizing monosyllable Chinese words using the method of vector quantization (VQ) and a hidden Markov model (HMM) is described. The theories of vector quantization and hidden Markov models are reviewed and studied the way they are used in recognizing monosyllable Chinese speech. Special attention was paid to sample program in order to obtain as much information as possible. A test was conducted on a small vocabulary containing thirty monosyllable Chinese words which have exactly the same compound vowel. The classification accuracy obtained is about 85% compared with the traditional DTW (decoder timing waveform), method which is not more than 60%