9th Annual Conference of the International Speech Communication Association

Brisbane, Australia
September 22-26, 2008

T-Tilt: A Modified Tilt Model for F0 Analysis and Synthesis in Tonal Languages

Ausdang Thangthai, Nattanun Thatphithakkul, Chai Wutiwiwatchai, Anocha Rugchatjaroen, Sittipong Saychum

NECTEC, Thailand

This paper proposes a modified Tilt model, called T-Tilt, for analyzing and synthesizing F0 contours in tonal languages. The Tilt model successfully designed for intonation modeling is extended to cover syllable-based F0 realization influenced strongly by the tonal context. Two modification approaches include adding a parameter indicating a F0 curve pattern and separating duration and amplitude controls inherent in the Tilt parameter for sake of flexibility. Evaluations are conducted by both an objective RMSE measure and a subjective MOS test on intelligibility and naturalness aspects. Applying to Thai and Mandarin Chinese continuous speech, the proposed model is proved to be very effective for F0 contour analysis. It rather requires extensive work on parameter synthesis although the synthesizing performance is comparable to those produced by other proposed models.

Full Paper

Bibliographic reference.  Thangthai, Ausdang / Thatphithakkul, Nattanun / Wutiwiwatchai, Chai / Rugchatjaroen, Anocha / Saychum, Sittipong (2008): "T-tilt: a modified tilt model for F0 analysis and synthesis in tonal languages", In INTERSPEECH-2008, 2270-2273.