基于音节韵律特征分类的汉语语音合成中韵律模型的研究  

Study of prosody model on Chinese speech synthesis based on the classification of syllabic prosody features

作　　者：陶建华[1] 蔡莲红[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《声学学报》2003年第5期395-402,共8页Acta Acustica

基　　金：863资助项目(2001AA114072)

摘　　要：论述了采用基于统计模型进行韵律建模的思路。在此基础上,提出了基于音节韵律特征分类的韵律建模思路,并采用韵律模板和韵律代价函数实现了韵律的自动预测。对该模型的自动训练算法进行了详细的阐述。根据统计的韵律建模方法,还分析了韵律特征间相互关联对音节韵律模板选取的影响。最后,进一步分析了统计韵律模型的进行韵律预测的误差分布情况,表明了该模型能够使语音合成系统具有较高自然度和高灵活性的特性。A prosody modeling method based on statistic model is described. Based on this, a Chinese prosody model based on the classification of syllabic prosody features is presented, which makes automatic prosody prediction with prosody templates and prosody cost function. And the automatic training algorithm of the model in detail is described. Further more, according to statistic prosody modeling method, the influence to prosody template selection with the help of the analysis of the prosody interaction among prosody elements is analyzed. Finally, the error distribution of the statistic method based prosody prediction is given. The results show good naturalness and much flexible in application.

关 键 词：统计模型 汉语 语音合成 韵律模型 韵律建模 音节韵律特征分类 韵律模板 韵律代价函数 

分 类 号：TN912.3[电子电信—通信与信息系统]