基于词类序列的汉语语句韵律结构预测  

Chinese sentence prosodic structure prediction based on the sequence of the parts of speech

作　　者：裴雨来[1,2] 邱金萍[2] 王洪君[1] 吕士楠[3] 

机构地区：[1]北京大学中文系,北京100871 [2]西安外国语大学,西安710128 [3]中国科学院声学研究所,北京100080

出　　处：《清华大学学报（自然科学版）》2009年第S1期1339-1343,1355,共6页Journal of Tsinghua University(Science and Technology)

基　　金：国家"八六三"高技术研究发展计划重点项目(2007AA01Z198)

摘　　要：为完善计算机汉语语音合成,构建了韵律结构预测系统。该系统区分了现代汉语中韵律结构与语法结构存在对应关系和不存在对应关系的两类单位,采取确定韵律结构和调整韵律结构,分别处理这两类单位;考虑了"的"、"介词"等功能词在韵律短语预测中的作用;面向汉语文语转换应用的语句韵律词及韵律短语;只依赖语句的词类音节数序列等简单的文本信息。在107句语料上进行了测试。结果表明:韵律词预测正确率达95%,韵律短语预测正确率约81%,操作性较强。A psosodic structure prediction system was built to improve the sound synthesization of Chinese sentences.This system distinguishes the grammar structures relating to the prosodic structure from the grammar structures not relating to the prosodic structure in modern Chinese,considering the effect of function words such as 'De' and some prepositions on predicting prosodic phrases.The prosodic structure of the Chinese sentence is then predicted based on the parts of speech and the sequence and number of syllables.Test using a corpus of 107 sentences show that the accuracy for predicting prosodic words is 95% and the accuracy for predicting of prosodic phrases is about 81%.The system can improve the quality of Chinese text to speech systems and can be easily operated.

关 键 词：语音合成 韵律语法 汉语文语转换 韵律结构 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]