To meet classic but still intrinsic requirements of speech synthesis in tonal languages, the tone nucleus model is employed in Thai. Tone nuclei are defined for all five distinctive tones according to the underlying targets. The full process of F 0 contour generation is presented from the nucleus extraction until the F 0 contour generation for the continuous speech. The objective and subjective tests confirm the efficiency and adaptability of the model in the other language than Mandarin. Compared to the F 0 contours generated by the predictors trained from the contours in the whole syllables without extracting the tone nuclei, the model generates the F 0 contours in the continuous utterances with less distortion from the target but more tone intelligibility and more naturalness. The modified method is also introduced to enhance the accuracy in the F 0 generation by the model.

No abstract