Using FO Contour Generation Process Model for Improved and Flexible Control of Prosodic Features in HMM-based Speech Synthesis

Authors: Keikichi Hirose, Keiko Ochi, Miaomiao Wang, Tatsuya Matsuda, Miaomiao Wen, Nobuaki Minematsu

Abstract:

Generation process model of fundamental frequency contours known as Fujisaki's model is ideal to represent global features of prosody. It is a command response model, where the commands have clear relations with linguistic and para/non linguistic information included in the utterance. Therefore, by controlling fundamental frequency contours in the framework of the generation process model, a more flexible control of prosodic features comes possible in speech synthesis. Also, the model can be used to solve the problems of HMM-based speech synthesis, which arise from frame-by-frame treatment of fundamental frequencies. In this paper, two methods for improved control of prosodic features in HMM-based speech synthesis, and one method for flexible fundamental frequency control to realize prosodic focuses in synthetic speech, are presented. All these methods are based on the generation process model.