一种汉语语音变换技术

汉语语音变换技术的目的是将汉语语音中源说话人的语音特征转换为目标说话人语音特征。提出的适用于汉语说话人的变换算法分为3个部分:前两部分用高斯混合模型实现了语音的谱包络(线性预测编码)及其激励(残差)的转换;第三部分采用支持向量回归算法实现语音的韵律变换规则建模,结合汉语语音特点利用基音同步叠加算法实现语音的超音段特征调整。与现有的语音变换算法进行比较,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变换并得到高自然度合成语音,是一种有效的汉语语音变换算法。

Chinese Speech Conversion Technology Based on LPC and Residual Model

Speech conversion is a new technology to change the source speaker's features to the target speaker's features in the speech. In this paper, the Chinese speech conversion system is divided in to three parts. In the first and second part, GMM(Gaussian Mixed Model) is used to transform the spectral envelopeLPC(Linear Prediction Coding)] and the impulse (residual). In the third part, the Chinese speech's super-segmental features is regulated with the SVR(Support Vector Regression) and the TD-PSOLA(Time-Domain Pitch Synchronous OverLap-Add). This algorithm is capital of transforming Chinese speech and producing spontaneous voice.