Modeling Vietnamese Speech Prosody: A Step-by-Step Approach Towards an Expressive Speech Synthesis System

Attempts to add expressivity to synthesized speech is one of the main strategies in speech technologies. This paper summarizes our researches on modeling Vietnamese prosody, with the goal of improving naturalness of synthesized speech in Vietnamese, as well as integrating expressivities (i.e. emotion/attitude). Based on the concept of "rendez-vous" between linguistic levels and prosodic functions, the prosody of utterance is proposed to be decomposed into several components. Therefore, each component is step by step modeled by an independent model: a dynamic linear segment model for tones, a relative registers model for F0 level of syllable, a rule-based approach for phrasing modeling and a F0 stylization modeling for the expressive function. All proposed models were integrated in speech Text-to-speech systems and also were evaluated by perception experiments.