基于混合基元模型的非定长基元选取算法  

Hybrid Unit Model Based Non-uniform Unit Selection

作　　者：倪昕[1] 蔡莲红[1] 

机构地区：[1]北京清华大学计算机系,北京100084

出　　处：《小型微型计算机系统》2005年第6期1079-1082,共4页Journal of Chinese Computer Systems

基　　金：国家"八六三"高技术项目(2002AA1170)资助;国家自然科学基金(60275014)资助

摘　　要：介绍了面向中英文双语应用的英文语音合成系统中基于混合基元模型的非定长基元选取算法.清华大学计算机系人机语音交互实验室针对中英文混读相同发音人的限定,实现了基于混合基元模型的语料库构建和鲁棒灵活的非定长基元选取方法,在一定程度上弥补了发音人英语发音不饱满、自由度大的缺陷,真正实现了相同发音人中英文混读的要求.试验证明,采用这些方法能够极大的提高合成质量,达到令人满意的效果.This paper presents hybrid unit model based non-uniform unit selection in Chinese and English bilingual application oriented English text-to-speech system. In order to meet the speaker-consistent requirement and address the nonnative speaker problem which manifests deficiency and unrestraint in pronunciation, the Human Computer Speech Interaction Laboratory at Tsinghua University constructs acoustic inventory based upon hybrid unit model and implements an efficient, robust and flexible non-uniform unit selection method. All these prove effective experimentally in dealing with the nonnative speaker problem and in promoting the overall synthesis quality.

关 键 词：文语转换(TTS) 混合基元模型 非定长基元选取 语料库 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]