Prosody Modeling of Mandarin Speech and its Applications

This presentation gives an overview of the recent studies on Mandarin-speech prosody modeling conducted in the Speech Processing Lab of NCTU. First, a prosody labeling and modeling method to automatically construct a hierarchical prosodic model (HPM) of Mandarin speech from a prosody-unlabeled speech corpus is discussed. Then, the method is extended to further consider the influences of speaking rate on both acoustic-prosodic features and HPM parameters. Last, three applications are introduced. One is a study to use the HPM to assist in Mandarin speech recognition. Another is the use of the HPM in Mandarin prosody coding. The other is the application to break prediction for Mandarin text-to-speech.