介绍说话人声音转换系统框架,并对传统的基于码本映射的说话人声音转换方法进行讨论.指出传统的码本映射方法由于对谱的转换采用所有码本加权叠加,因此会产生转换后语音频谱平滑效应过重的问题,从而使转换后语音音质较差.为了克服这种问题,本文提出基于音素绑定的码本加权叠加方法来完成语音谱的转换,同时利用决策树来完成韵律的转换.实验表明,即使在数据量较少的情况下,该方法也能较好地完成说话人声音转换,并能得到较高的语音音质. This paper introduces the framework of the speaker’s voice conversion system and discusses the speaker’s voice conversion method based on the traditional codebook mapping.It is pointed out that the traditional codebook mapping method uses all the codebook weighted overlays for spectral conversion and thus generates converted speech In order to overcome this kind of problem, this paper proposes a codebook weighting superposition method based on phoneme binding to complete the conversion of speech spectrum, at the same time the decision tree is used to complete the prosody The experiment shows that this method can accomplish the speaker’s voice conversion better and get higher voice quality even in a small amount of data.