文语转换系统中的韵律研究

Study on the Prosody in Text-to-Speech Systems

1 引言 文语转换(Text-To-Speech,TTS)是将文字形式的信息转换成自然语音的一种技术,在人机交互、通讯、资讯、家电等领域有着广泛的用途.然而,当前的TTS系统普遍存在着输出语音的机器味太浓、不够自然的现象,在很大程度上阻碍了它的推广和应用.其根本原因即在于合成语音中缺乏必要的韵律信息.我们认为,一个TTS系统应分为相对独立的上下两层:韵律结构分析和语音生成.上层负责分析语句的韵律结构,并标注相应的韵律标记,下层负责将之转换成相应的合成器参数,并输出语音.因此,当前的首要任务就是要研究韵律的主要特点、韵律的结构和主要内容,在此基础上,制订出一套相应的韵律标记方法.

The paper discussed the problem of prosody in Text-to-Speech(TTS) systems. It firstly analyzed the different markup methods employed in current TTS systems ,then discussed the characteristics of prosody in detail,and put forward a comprehensive and integrated Chinese prosodic strcture. Lastly a new markup language named Chinese Prosodic Markup Language(CPML) was proposed. The features of CPML are :1. Abstract and platform-independent, 2. Covering most of important prosodic information, 3.Hierarchical prosodic structure ,and 4.Open and updating.