8th International Conference on Spoken Language Processing

Jeju Island, Korea
October 4-8, 2004

Formulating Contextual Tonal Variations in Mandarin

Jinfu Ni (1), Hisashi Kawai (1), Keikichi Hirose (2)

(1) ATR, Japan
(2) University of Tokyo, Japan

This paper deals with both tone modeling and contextual tonal variations for formulating the latter in a parametric form based on a functional fundamental frequency (F0) model.The contextual tonal variations are measured in terms of F0 peak and valley targets from 1,560 Chinese polysyllabic words, taking into account two factors of contextual tones and target tone position. The speaker-dependent effect on these observed samples is eliminated by voice range normalization. In the formulation with tone modeling, the peak target of a tone is represented as its bias with respect to a natural declination line, while the gliding feature of the tone is expressed as the response time and amplitude of transition between its valley and peak targets. This paper presents the details of formulated parameters for tri-tone contexts and algorithms for converting tonal peak and valley targets into the model parameters for synthesizing F0 contours. The achievement is useful for synthesizing neutral intonation.

Full Paper

Bibliographic reference.  Ni, Jinfu / Kawai, Hisashi / Hirose, Keikichi (2004): "Formulating contextual tonal variations in Mandarin", In INTERSPEECH-2004, 749-752.