A method for modeling and generating Mandarin tone contour with phrase intonation based on the generation process model

This paper models F₀ curves as the representation of both syllable-level tone and phrase-level intonation based on generation process model for Chinese Mandarin speech. The tone components are realized by concatenating their fragments predicted by a HMM-based method, while the phrase components are generated by rules under the generation process model framework. In the conventional HMM-based TTS, intonation, especially at the sentence or phrase level, tends to be bland. The Multi-Space Distribution (MSD) used in HMM training and F₀ parameter trajectory generation is partially responsible for the blandness. Additionally, the F₀ trajectory thus generated has a smaller dynamic range than that of natural speech, and makes the synthesized speech sound less lively. In this paper, in order to model the F₀ within the standard HMM framework, an F₀ generation process model is used to re-estimate the tone components and phrase components. A prior knowledge of VU is imposed in each Mandarin phoneme and they are used for VU decision. Also we design a set of syntax features to improve Mandarin phoneme duration prediction.