一种用于语声转换系统的LPC残差信号生成算法  

An Algorithm for Generating LPC Residual Signal Used in Voice Conversion System

作　　者：简志华[1] 杨震[1] 

机构地区：[1]南京邮电大学信号处理与传输研究院,江苏南京210003

出　　处：《信号处理》2008年第5期762-765,共4页Journal of Signal Processing

基　　金：江苏省青蓝工程项目资助;编号QL003YZ

摘　　要：残差信号中的基音信息对语音的说话人个性特征有着重要的影响。本文首先通过转换后的语音谱包络特征参数(LSP)来预测相应的目标基音周期,再利用预测的目标基音周期来修改源语音的残差信号,从而生成所需要的目标语音残差信号。客观评测和主观听觉测试都表明,本文的残差信号生成算法(PP+DCT,Pitch Prediction+Discrete Cosine Trans- form)性能要好于以往的残差预测法。Pitch information contained in residual signal of speech plays an important role in speaker identity. An algorithm （ PP ＋ DCT, Pitch Prediction ＋ Discrete Cosine Transform） for generating residual signal is proposed in this paper,which firstly predicts the corresponding target pitch using the converted spectral envelop feature （LSP） , and then uses DCT to modify the source residual signal according to the dynamic pitch - scale factor. Both subjective and objective evaluations are conducted. The experimental results demonstrated that the proposed algorithm can achieve better performance than the residual prediction （RP） method.

关 键 词：语音处理 语声转换 残差信号 基音预测 离散余弦变换 

分 类 号：TN713[电子电信—电路与系统] TP391.41[自动化与计算机技术—计算机应用技术]