An HMM-based Cantonese speech synthesis system

This paper describes a Cantonese HMM-based speech synthesis system (HTS) using the general architecture of Crystal - a multilingual text-to-speech (TTS) framework developed in Tsinghua University. The generated synthesis engine of HTS has advantage of small footprint, the size of which is less than 7M bytes, and can be easily ported to embedded electronic devices such as smart-phones, set-top boxes, etc. Furthermore, the quality of the synthetic speech can be easily characterized by modifying the synthetic acoustic parameters of the proposed system. The result shows noticeable improvement in naturalness and smoother transition than the corpus-based unit-selection concatenative speech synthesis approach.