International Symposium on Chinese Spoken Language Processing (ISCSLP 2002)

Taipei, Taiwan
August 23-24, 2002

Hakka Pitch-Contour Parameter Generation Using a Mandarin-Trained Pitch-Contour Model

Hung-Yan Gu, Shiue-Jen Li

Department of Computer Science and Information Engineering, National Taiwan University of Science and Technology, Taipei

In this paper, using an existing pitch-contour model of a Chinese dialect (Mandarin here) to generate pitch-contour parameters for another dialectís sentences (Hakka here) is studied. This can be generally viewed as a pitch-contour model adaptation problem. We study this problem in hope to save tedious labors and research time needed to build a pitchcontour model for a specific Chinese dialect. This approach is also useful for synthesizing speech of a weak dialect to help reserve it from disappearance. In this study, we have built a prototype Hakka speech synthesis system. Except the pitchcontour parameters, the other prosodic parameters are generated by rules. For signal-waveform synthesis, TIPW method previously proposed is adopted. Two sets of Hakka and Min-Nan sentences are synthesized, respectively, for evaluation experiments. Syllable signals for Min-Nan are borrowed from the recorded Hakka syllables, and pitchcontour parameters are generated using the same approach. Because of timing difference and different difficulties in sentence contents, the comprehension and naturalness-level scores for Hakka synthetic speech only reach 91.87% and 79.5%, respectively. But for the set of Min-Nan synthetic speech, the scores obtained are 97.1% and 85.5%, respectively.


Full Paper

Bibliographic reference.  Gu, Hung-Yan / Li, Shiue-Jen (2002): "Hakka pitch-contour parameter generation using a Mandarin-trained pitch-contour model", In ISCSLP 2002, paper 49.