A trainable Vietnamese speech synthesis system based on HMM

摘要

This paper describes an approach to the realization of a trainable speech synthesis system using a technique whereby speech is directly synthesized from Hidden Markov models (HMMs), and apply it to the Vietnamese synthesis system. A series of data should be prepared before the training and synthesis process, including the collection of data, recording, labelling, the design of contextual property and problem sets. The whole training and synthesis process is based on the HMM-based Speech Synthesis System-2.0 (HTS-2.0) and it is automated. The final synthesis result shows that using this method to the Vietnamese synthesis system is feasible.