基于自适应粒子群优化径向基函数神经网络的语音转换  

Voice Conversion Based on Adaptive Particle Swarm Optimization Radial Basis Function Neural Network

作　　者：张玲华[1] 姚绍芹 解伟超[2] 

机构地区：[1]南京邮电大学物联网学院,南京210003 [2]南京邮电大学通信与信息工程学院,南京210003

出　　处：《数据采集与处理》2015年第2期336-343,共8页Journal of Data Acquisition and Processing

基　　金：江苏省高校自然科学研究重大(13KJA510003)资助项目;江苏高校优势学科建设工程(PAPD)资助项目;江苏省普通高校研究生科研创新计划(CXLX12_0478)资助项目

摘　　要：语音转换是指在保持源说话人语义内容不变的前提下,通过改变源说话人的个性特征,使其听起来像目标说话人的语音。本文提出一种自适应粒子群优化算法训练径向基函数神经网络进行语音特征建模,以获取说话人谱包络的映射关系;此外,考虑到说话人谱包络参数与基频有着密切的联系,利用基于径向基函数神经网络的联合谱包络基频变换方法,将谱包络参数与基频联合进行建模和转换,使得转换后的基频含有更多的说话人个性特征。最后,运用主、客观方法对获得的转换语音进行性能测试。实验表明,与主流的基于高斯混合模型的语音转换相比,使用自适应粒子群优化的径向基函数神经网络方法能够获得更好的转换性能,且更加适用于男声到女声的转换。Voice conversion is a technique for changing the personality characteristics of a source speaker′s voice into the target speaker′s,while preserving the original semantic information.An adaptive particle swarm optimization（PSO）based method is proposed to model voice features by training the radial basis function（RBF）neural network in order to capture the spectral envelope mapping function between speakers.In addition,the pitch transformation is captured by modeling pitch with the joint spectral feature parameters in RBF neural network,which makes the converted pitch contain more target details.Finally,the performance of the improved voice conversion system is tested by subjective and objective method respectively.Experimental results show that the performance of the proposed method is better than that of the Gaussian mixture model（GMM）based system,especially for the male to female conversion.

关 键 词：语音转换 径向基函数神经网络 自适应粒子群优化 高斯混合模型 基频 

分 类 号：TN912.3[电子电信—通信与信息系统]