基于HMM的英语文语转换系统

HMM-Based English Text-to-Speech System

研究了基于隐马尔可夫模型(HMM:Hidden Markov Model)的语音合成系统的关键技术,在此基础上,借助HTK和Festival等工具,以基频和声道谱参数为训练参数,实现了一个基于HMM的英语文语转换系统,主观试听,合成的语音流畅、清晰可懂,并把混合激励应用到系统中对激励进行改进,提高了自然度。实验结果表明,利用HMM技术实现合成单元的选择,较好地解决了文语转换系统中的协同发音的问题。

This paper focuses research on the key technology of the speech synthesis system based on HMM （Hidden Markov Model） , and, with the help of HTK and Festival, implements an English TTS system taking as training parameters the fundamental frequency and the tract parameter. The synthesized voice turns out to be clear, smooth and understandable. Applying mixed excitation to system, speech of the output is more natural. Experiments show that the application of HMM in the selection of synthetic unit can effectively solve the problem of co-articulation in text to speech systems.