ISCA Archive ISCSLP 2002
ISCA Archive ISCSLP 2002

Annotation of Chinese prosodic level based on probabilistic model

Rui Cai, Zhi-Yong Wu, Lian-Hong Cai

In this paper, a probability based method is proposed for the annotation of Chinese prosodic levels. We investigated the acoustic correlates (F0 pitches, durations and pauses) of the prosodic boundaries and tried to annotate the levels of the boundaries using Gaussian model and Bayesian decision. The method allows efficient and automatic labeling for the large scale speech corpus and can be used to instruct the unit selection in TTS system.