基于多情感说话人自适应的情感语音合成研究  

Research on Emotion Speech Synthesis Based on Multi Emotional Speaker Adaptive

作　　者：李葵 徐海青 吴立刚 梁翀 LI Kui;XU Hai-qing;WU Li-gang;LIANG Chong(Information ＆ Communication Branch,State Grid Anhui Electric Power Company,Hefei 230000;School of Computer Science,Xiamen University,Xiamen 361005;Anhui Jiyuan Software Co.,Ltd.,Hefei 230000 China)

机构地区：[1]国网安徽省电力有限公司信息通信分公司,安徽合肥230000 [2]厦门大学计算机科学学院,福建厦门361005 [3]安徽继远软件有限公司,安徽合肥230000

出　　处：《湘潭大学学报（自然科学版）》2018年第4期39-44,共6页Journal of Xiangtan University(Natural Science Edition)

基　　金：国网总部科技项目(5268001600SV)

摘　　要：提出了一种基于多情感自适应的情感语音合成方法,其创新点在于,通过SAT过程从多个说话人的情感语音语料中获得情感语音的平均音模型,对目标说话人的情感数据进行自适应变换,构建目标情感的声学参数模型,从而达到合成出目标说话人的情感语音的效果.实验表明,本方提出的方法能够获得自然度和情感相似度均较好的合成情感语音.In this paper, an emotional speech synthesis method based on multi emotion speaker adaptation is proposed, which distinguishes the traditional speech synthesis algorithm, and uses multiple speaker’s emotional speech materials to obtain the average voice model of multiple speaker’s emotional voice through SAT. The target speaker’s emotion data is adaptively transformed to construct, the acoustic model of target emotion, and then the target speaker’s target emotion speech is synthesized. Experiments show that the synthesized emotion speech has high naturalness and emotional similarity.

关 键 词：情感语音合成 情感语料库 SAT 统计参数语音合成 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]