Synthesizing intonation for Malayalam TTS

In this paper we propose a new method for modeling intonation for Malayalam language Text To Speech (TTS) synthesis system. It is obtained by combining curve fitting and CART. Intonation provides naturalness and intelligibility to synthesized speech. Intonation of a sentence is predicted from raw text using positional information of syllables and words. The intonation contour of first word is curve fitted for enhancing performance of CART model. Intonation of synthesized speech is varied dynamically according to predicted intonation contour of sentence. Statistical evaluation of model is conducted.