维吾尔语音情感声学特征提取与建模研究  

Acoustical Feature Extraction and Modeling of Emotional Uyghur Speech

作　　者：阿依提拉.米吉提 艾斯卡尔.肉孜 艾斯卡尔.艾木都拉[1] 

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046 [2]新疆大学数学与系统科学学院,新疆乌鲁木齐830046

出　　处：《通信技术》2013年第11期51-54,共4页Communications Technology

基　　金：国家自然科学基金资助项目(No.61065005);教育部新世纪优秀人才支持计划项目(No.NCET-10-0969)~~

摘　　要：在收集较大规模情感语音语料库基础上,分析了维吾尔语语音在韵律特征和音质特征方面的19种语境信息和6种情感特征参数,并利用STRAIGHT算法实现了情感特征参数的提取,最后利用分类回归树(CART)算法针对各个情感特征和中性向其它情感的转换特征进行了建模。实验结果表明,所提取情感特征能准确的区分各个情感类型,为实现中性语音转换成各种目标情感语音奠定了基础。Based on large-scale emotional Uyghur speech database, this paper analyzes 19 contextual infor- mations and 6 emotional prosodic features of the speech data. STRAIGHT algorithm is used to implement the extraction of emotional characteristic parameters, and finally the classification and regression tree （CART） algorithm is used to build up the individual models for each kind of emotional speech and models for emo- tional conversion from neutral to another given emotion. Experimental results show that the extracted emo- tional characteristics could be used to accurately distinguish various types of emotions. All this would lay a foundation for achieving the conversion of neutral speeches to various target emotional speeches.

关 键 词：维吾尔语 情感特征 STRAIGHT 算法 韵律分析 情感建模 

分 类 号：TN912[电子电信—通信与信息系统]