基于决策树的多特征语音情感识别  

Research of Speech Emotion Recognition Based on Decision Tree and Acoustic Features

作　　者：石瑛[1] 胡学钢[1] 方磊[1] 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《计算机技术与发展》2009年第1期147-149,175,共4页Computer Technology and Development

基　　金：国家自然科学基金(10605022);安徽省自然科学基金(050420207)

摘　　要：数据挖掘技术是近年来计算机领域的重要方向。文中的研究目的就是通过深入分析各种语音情感特征,找出某种特征对语音情感识别的贡献程度,并在数据挖掘技术中寻找适合的模型将有效特征加以利用。分析和研究了多位科学家在进行语音情感分析过程中采用的方法和技术,通过总结和创新建立了语音情感语料库,并成功地提取了相关的语音信号的特征。后研究了基音频率、振幅能量和共振峰等目前常用的情感特征在语音情感识别中的作用,把数据挖掘中常用的决策树分类方法和语音信号的多个特征相结合,建立了语音情感识别模型,对语音情感数据进行了大量的实验,取得了较为满意的识别结果。Data mining is one of the important computer technologies. The main goal of this thesis is to search the most useful features with analyzing the features related with emotions, and find a recognition model to make use of thee features. Studied the method and technology in the research of the speech emotion recognition. Created the database of the speech emotion recognition and pick- up the features of the speech signal. Then study the effect in emotion-speech recognition from those common features such as pitch, amplitude energy, formant and so on. And also use decision tree with multi - features to recognize speech emotion, the result is perfect.

关 键 词：语音情感识别 情感特征 数据挖掘 决策树 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]