基于SVR的语音变换映射规则研究  

The Algorithm of Voice Transformation Based on SVR

作　　者：崔丽珍[1] 宋巍[2] 

机构地区：[1]内蒙古科技大学信息工程学院,包头014010 [2]解放军理工大学通信工程学院,南京210007

出　　处：《电子与信息学报》2008年第5期1144-1147,共4页Journal of Electronics & Information Technology

摘　　要：该文介绍了语音变换与支持向量回归(SVR)的基本理论。提出了基于多输出支持向量回归的语音变换特征参数映射规则,并对该映射规则进行了仿真实验。对变换后语音所进行的主客观测试表明,该映射规则对比码书映射和高斯混合模型,能够在参数映射离散性和平滑性之间有效折中,提高语音可懂度。In this paper, the fundamental theory of voice transformation and Support Vector Regression（SVR） are introduced. The algorithm of voice transformation technology based on multi-output SVR is put forward, and the corresponding simulation experiment is carried out. The subjective test and objective test show that compared with code book mapping and GMM, this mapping rule performs well between discreteness and smoothness of parameter.

关 键 词：语音变换 映射规则 支持向量回归 

分 类 号：TN912.3[电子电信—通信与信息系统]