基于STRAIGHT模型的语音转换系统研究

Research on Speech Conversion System Based on STRAIGHT Model

语音转换是将源说话人的个性特征转换为目标说话人个性特征的过程。主要研究了基于STRAIGHT模型的语音转换系统原理及实现过程。通过STRAIGHT模型提取目标语音和源语音的基本频率以及平滑的声道频谱作为特征参数,并将声道频谱转换为LSF参数,进行时间对齐和GMM训练。从实验结果数据分析可以看出:由STRAIGHT模型提取的参数很好地避免了声道谱过平滑的现象,合成后的目标语音与源语音的相似度较高。

Speech conversion is the process of transforming the personality characteristics of the source speaker into the personality characteristics of the target speaker.This paper mainly studies the principle and implementation process of speech conversion system based on STRAIGHT model.The STRAIGHT model is used to extract the basic frequency and smooth channel spectrum of target and source speech as feature parameters,and the channel spectrum is converted into LSF parameters for time alignment and GMM training.The data analysis of the experimental results shows that the parameters extracted by the STRAIGHT model can avoid the phenomenon of too smooth channel spectrum,and the synthesized target speech has a high similarity with the source speech.