分层语音合成数据库的分析与裁减  

Analysis and tailor of layered corpus for speech synthesis

作　　者：易立夫[1] 孙金城[1] 田静[1] 

机构地区：[1]中国科学院声学研究所,北京100080

出　　处：《声学学报》2004年第2期167-170,共4页Acta Acustica

摘　　要：提出汉语分层语音合成数据库的概念,通过对其分析和裁减,有如下结果:(1)数据库音节基频、时长、能量的变化有层次.(2)数据库的拼接合成质量接近大语料库的拼接合成质量。(3)数据库为实现表情语调的拼接合成奠定了基础。(4)数据库资源消耗远小于大语料数据库的资源消耗。The concept of Chinese layered corpus for Chinese speech synthesis has been brought forward in this paper. Through the analyzing and tailoring of the layered corpus, some important conclusions are obtained: (1)The pitch, duration and energy of corpus are layered. (2) The naturalness of this layered corpus can compare with the large corpus for concatenative speech synthesis. (3) The achievement of the expressive intonations for concatenative speech synthesis may be based on the layered corpus. (4) The use of resources of layered corpus is much less than the large corpus.

关 键 词：分层语音合成数据库 拼接合成质量 语调拼接 资源消耗 汉语语音合成 表情语调 中性语调 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]