采用谱包络与超音段韵律调整的高自然度语音转换  

Voice Conversion with High Naturalness Using Spectrum and Super-segmental Feature Transform

作　　者：丁耀娥[1] 俞一彪[1] 

机构地区：[1]苏州大学电子信息学院,江苏苏州215021

出　　处：《苏州大学学报（工科版）》2009年第4期10-15,共6页Journal of Soochow University Engineering Science Edition (Bimonthly)

基　　金：国家自然科学基金资助项目(编号60572076)

摘　　要：汉语语音转换技术的目的是在保持语义信息不变的前提下将汉语语音中源说话人的语音特征转换为目标说话人的语音特征。提出的适用于汉语说话人高自然度语音转换算法主要分为三个部分:前两部分用VQ模型实现了语音的谱包络及其激励的转换,第三部分采用BP算法实现语音的韵律转换规则建模。算法针对汉语语音超音段发音特点进行韵律调整,有效实现了汉语语音转换并得到高自然度合成语音。实验表明本文提出的转换方法是一种有效的汉语语音转换算法。Voice conversion is a technology to convert the individuality of source speaker＇s speech to the target speaker＇s. In this paper,a new Chinese voice conversion system is proposed which includes following three parts：spectral envelop transform; exciting source transform; super-segmental prosodic adjustment. In the first and second parts, vector quantization is used to transform the spectral envelope and the exciting（residual） from source speaker＇s voice to target speaker＇s voice. In the third part, the super-segmental feature of speech is regulated network. Experiments show proposed algorithm is effect for voice conversion and producing with the back-propagation quite natural voice.

关 键 词：语音转换 VQ模型 超音段特征 BP网络 

分 类 号：TP912.3[自动化与计算机技术]