KD2000 Chinese Text-To-Speech System

Abstract

This paper presents the new progress we made in recent two years on Chinese text-to-speech towards higher naturalness. The results can be summarized as follows: 1). Aim at the different characteristics between philology and phonetics, a kind of hierarchy process model for Chinese TTS system has been proposed. Five layers are defined to label the sentence in the text analysis. 2). A prosodic generating model is built for selecting appropriate unit with higher accuracy. 3). Extracting prosodic parameter from the unit base and then normalizing them to form a prosodic parameter base. 4). Proposed an effective method for processing the special text, which integrates external descriptor rules and model matching. With these progresses a new Chinese text- to-speech system named KD2000 has been developed. The improved performance has been confirmed by evaluation test.