Affiliation:
1. Hankou University
2. Wuhan Digital Engineering Institue
3. Synopsys Inc.
Abstract
After about 50 years of development, speech recognition technology has been able to achieve large vocabulary, non-specific human continuous speech recognition system. On account of Chinese pronunciation features, we research the small vocabulary, non-specific Chinese speech recognition based on continuous Hidden Markov Model approach. With comparing the datasets of VQ/DTW, VQ/DHMM, CHMM state-1 recognition algorithm and CHMM state-2 recognition algorithm, the results of our experiment show that: (1) CHMM state-2 branch method performs primely in reduction of the recognition time; and (2) the recognition accuracy is improved eventually.
Publisher
Trans Tech Publications, Ltd.
Reference9 articles.
1. Myers C S, Rabiner L R, Rosenberg A E. An Investigation of the Use of Dynamic Time Warping for Word Spotting and Connected Word Recognition. ICASSP 80, pp.173-177.
2. Higgins Alan L, Wohlord Robert E. Keyword Recognition Using Template Concatenation. ICASSP85, 1985(3) pp.233-1236.
3. Wilson J G, Lee C H, Rabiner L R. Application of Hidden Markov Model for Recognation of a Limited Set of Words in Unconstrained Speech. ICASSP85, 1989(3) pp.173-177.
4. Wilson Jay G, Rabiner Lawrence R, etal. Automatic Recognation of Key-words in Unconstrained Speech Using Hidden Markov Model. on ASSP, 1990, 38(11) p.1980-(1978).
5. A. Martin，GDoddington，T. Kamm，M. ordowski，andM. Przyboeki. The DET Curve in Assessment of Detection Task Performance. In Proc. Eurospeeeh. Rhodes，Greece. 1997, pp.1895-1898.