8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


Corpus-Based Synthesis of Fundamental Frequency Contours of Japanese Using Automatically-Generated Prosodic Corpus and Generation Process Model

Keikichi Hirose, Takayuki Ono, Nobuaki Minematsu

University of Tokyo, Japan

We have been developing corpus-based synthesis of fundamental frequency (F_0) contours for Japanese. Since, in our method, the synthesis is done under the constraint of F_0 contour generation process model, a rather good quality is still kept even if the prediction process is done poorly. Although it was already shown that the synthesized F_0 contours sounded as highly natural as those using heuristic rules carefully arranged by experts, the F_0 model parameters for the training corpus were extracted with some manual processes. In the current paper, the automatically extracted parameters are used, and a good result is obtained. Also several features are added as the inputs to the statistical method to obtain better results. Some results on the accent phrase boundary prediction in the similar corpus-based framework are also shown.

Full Paper

Bibliographic reference.  Hirose, Keikichi / Ono, Takayuki / Minematsu, Nobuaki (2003): "Corpus-based synthesis of fundamental frequency contours of Japanese using automatically-generated prosodic corpus and generation process model", In EUROSPEECH-2003, 333-336.