A New Prosodic Strength Calculation Method for Prosody Reduction Modeling

To improve the naturalness of synthetic speech, prosody models in text-to-speech (TTS) system should be able to describe different prosody variations in natural speech. In this paper, prosody variation patterns behind the partial reduction phenomena are analyzed. In order to model the prosody reduction effect and incorporate it into the prosody model for speech synthesis, prosodic strength is introduced and a new prosodic strength calculation method is proposed. The method aims to model the sentence planning of prosody reduction and is based on the concept that the objective of prosodic strength should complete the planned target of the speech unit. The approach on how to integrate prosodic strength into speech synthesis system is also introduced. Experiments show that the estimated prosodic strength values by the proposed method have good correlations with both prosody structure and acoustic features.