Title :
Automatic speech segmentation for Chinese speech database based on HMM
Author :
Tao, Jianhua ; Hain, Horst-Udo
Author_Institution :
Dept. of Comput. Sci. & Technol., Tsinghua Univ., Beijing, China
Abstract :
The paper offers an optimized method for speech segmentation of a Mandarin speech database by using a hidden Markov model (HMM). The method takes the syllable boundaries into account. Testing shows that the accuracy of results is improved to 95% from 88% compared to the normal method. In particular, most of the boundaries between two vowels can also be well detected with the new method. The paper also analyzes the influence of the amount of HMM states and the amount of the training corpus.
Keywords :
hidden Markov models; optimisation; speech processing; Chinese speech database; HMM states; Mandarin speech database; automatic speech segmentation; hidden Markov model; optimized method; syllable boundaries; training corpus; Computer science; Databases; Hidden Markov models; Labeling; Optimization methods; Paper technology; Personal digital assistants; Speech processing; Speech synthesis; Testing;
Conference_Titel :
TENCON '02. Proceedings. 2002 IEEE Region 10 Conference on Computers, Communications, Control and Power Engineering
Print_ISBN :
0-7803-7490-8
DOI :
10.1109/TENCON.2002.1181318
