Analysis of Chinese Interrogative Intonation and its Synthesis in HMM-Based Synthesis System

Abstract

In the training process of HMM-based speech synthesis systems, the states of the HMM models are clustered based on the context features labeled on the training corpus. A good feature set would enable the synthesis system to generate natural prosody. In this paper, in order to synthesize interrogative intonation, we analyzed the acoustic features of the interrogative intonation and the context features that are related to the interrogative intonation in Chinese. We then expanded the feature set used conventional HMM-based speech synthesis system with interrogative intonation related features. A new speech synthesis system is trained with the proposed feature set. The analysis of train and synthesis results show that the proposed feature set is able to discriminate declarative and interrogative intonations. Perception listening test also shows the synthesis system using the expanded feature set can generate interrogative intonation.