15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Syntactic Structure to Prosodic Structure Mapping with Inductive Learning Method

Jianhua Tao (1), Shen Zhao (2)

(1) Chinese Academy of Sciences, China
(2) Tsinghua University, China

During the last several years, there has been a rapid progress in Chinese speech synthesis. Now, the method of unit selection and concatenation, accompanying with large corpus, is used widely in the systems design. Nevertheless, the prosodic structure was still proved to be the essential links between linguistics and acoustics, and behaves as an important parameter for prosody processing and unit selection. Features related to prosodic boundaries are extracted with the corresponding boundary labels to establish a templates set. A transformational learning based method is applied to establish the prediction model for the prosodic phrasing and prosodic word parsing. The paper generates general evaluation parameters for prediction model. The importance of the features related to the prosody boundaries are also analyzed in the paper. The experiments show that the method approach can achieve the accuracy rate of 84% for prosodic word boundary and 76% for prosodic phrase boundary.

Full Paper

Bibliographic reference.  Tao, Jianhua / Zhao, Shen (2003): "Syntactic structure to prosodic structure mapping with inductive learning method", In ICPhS-15, 1145-1148.