Speech Prosody 2004

Nara, Japan
March 23-26, 2004

Duration Modeling for Mandarin Speech Recognition Using Prosodic Information

Wern-Jun Wang (1), Chun-Jen Lee (2,1)

(1) Internet & Multimedia Application Technology Laboratory, Chunghwa Telecom. Laboratories, Taiwan
(2) Department of Computer Science, National Tsing Hua University, Taiwan

In this paper, a new duration modeling method for HMMbased Mandarin base-syllable recognition is proposed. It extends the conventional state duration method to further consider the speaking rate of utterance and add a syllable duration model to help the recognition search finding the bestrecognized base-syllable string. Experimental results showed that the proposed method was effective on improving the recognition accuracy.

Full Paper

Bibliographic reference.  Wang, Wern-Jun / Lee, Chun-Jen (2004): "Duration modeling for Mandarin speech recognition using prosodic information", In SP-2004, 591-594.