Overview of the IBM Mandarin Text-to-Speech System

This paper presents overview of the IBM M andarin Text-to-Speech system. It is a conc atenative speech synthesis system with a da ta-driven text processing module and probability-ba sed prosody model. The TC-STAR evaluation results showed that the system is sta te-of-the-art both in intelligibility and naturalness, and the synthesis speech are close to natural speech in overall quality. The system also has the capability to fast-develop new voices, languag es, and Chinese dialects by using data-driven methods.