DocumentCode :
1607667
Title :
Entering tone recognition with hybrid method
Author :
Liu, Ying ; Wang, Xiangcheng
Author_Institution :
Dept. of Chinese Language & Literature, Tsinghua Univ., Beijing, China
fYear :
2010
Firstpage :
362
Lastpage :
366
Abstract :
Entering-tones are the tones which corresponding Chinese characters are ancient entering-tone characters. The entering tones are recognized using hybrid method which combines rule-based method with statistical method. The syllable which may possibly be an entering-tone or a non-entering tone is called ambiguity syllable. The non-ambiguity syllables can be recognized using rule-based method. The ambiguity syllables could be disambiguated with statistical method and the statistical model is maximum entropy model. The recognition accuracy using rules is 64.46%, the recognition accuracy using maximum entropy model is 91.61%, and the accuracy using hybrid method is 97.02%.
Keywords :
knowledge based systems; maximum entropy methods; speech recognition; statistical analysis; ambiguity syllable; entering tone recognition; entering-tone characters; hybrid method; maximum entropy model; rule-based method; statistical method; Accuracy; Character recognition; Context; Entropy; Mathematical model; Statistical analysis; Training; Entering Tone Recognition; Maximum Entropy Model; Rule;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Universal Communication Symposium (IUCS), 2010 4th International
Conference_Location :
Beijing
Print_ISBN :
978-1-4244-7821-7
Type :
conf
DOI :
10.1109/IUCS.2010.5666247
Filename :
5666247
Link To Document :
بازگشت