A New Prosodic Phrasing Model for Chinese TTS Systems

This paper proposes a new prosodic phrasing model for Chinese text-tospeech systems. First, in contrast to the commonly used CART techniques, we propose a new inductive learning algorithm based on the extension matrix theory. Second, we collected 559 sentences (of approximately 78 min length) from news programs and built a corresponding speech corpus uttered by a professional male announcer. The prosodic boundaries were manually marked and word identification, POS tagging and syntactic analysis were also done on the text. Finally, our model was trained on 371 sentences and tested on 188. We achieved a success rate of about 93%.

• Type: Conference
• Year: 2001
• Where: NLPRS
• Authors: Weijun Chen, Fuzong Lin, Jianmin Li, Bo Zhang
• Keywords: Extension Matrix Theory, Inductive Learning Algorithm, Natural Language Processing, NLPRS 2001, Prosodic Phrasing Model