International Symposium on Tonal Aspects of Languages
With Emphasis on Tone Languages

Beijing, China
March 28-31, 2004

Large Vocabulary Mandarin Chinese Continuous Speech Recognition System Based on Tonal Triphone

Long Yan, Rencai Zhao, Gang Liu, Jun Guo

School of Information Engineering, Beijing University of Posts and Telecommunications, Beijing, China

Large vocabulary mandarin Chinese continuous speech recognition has been a difficult problem in speech recognition area because of several reasons. First, it is a tone language. There are five lexical tones that are important in distinguishing the confusable words in mandarin. So the modeling of tones plays an important role in mandarin speech recognition. Second, the variation of tones in spontaneous mandarin speech would have some effects on the performance. Third, the co-articulation is inevitable in spontaneous mandarin speech recognition. In this paper, a large vocabulary mandarin Chinese continuous system based on tonal triphone was constructed. The experimental results shows that a good performance in acoustic level has been achieved while poor performance in word level.

Full Paper

Bibliographic reference.  Yan, Long / Zhao, Rencai / Liu, Gang / Guo, Jun (2004): "Large vocabulary Mandarin Chinese continuous speech recognition system based on tonal triphone", In TAL-2004, 221-224.