多重分形在语音情感识别中的研究  

Speech emotion recognition based on multifractal

作　　者：叶吉祥[1,2] 王聪慧[1] 

机构地区：[1]长沙理工大学计算机与通信工程学院,长沙410114 [2]中南大学信息科学与工程学院,长沙410083

出　　处：《计算机工程与应用》2012年第13期186-189,205,共5页Computer Engineering and Applications

基　　金：湖南省自然科学基金重点项目(No.10jj2050)

摘　　要：为了克服语音情感线性参数在刻画不同情感类型特征上的不足,将多重分形理论引入语音情感识别中来,通过分析不同语音情感状态下的多重分形特征,提取多重分形谱参数和广义Hurst指数作为新的语音情感特征参数,并结合传统语音声学特征采用支持向量机(SVM)进行语音情感识别。实验结果表明,通过非线性参数的介入,与仅使用传统语音线性特征的识别方法相比,识别系统的准确率和稳定性得到有效提高,因此为语音情感识别提供了一个新的思路。In order to overcome the inadequate of emotional conventional linear argument at depicting different types of character sentiments, this paper takes the multiple fractals theory into the sound emotional identify, by analyzing the multiple fractal features on the different sound emotional state, and proposes multifractal spectrum parameters and generalizes hurst index as emotional conventional parameters, combines with traditional voice acoustic features and using Support Vector Machine （SVM） for speech emotion recognition. The results show that the accuracy and stability of the recognition system are improved effectively through using non-linear parameters, compared with the linear features of traditional voice recognition method, so it provides a new idea for voice emotion recognition.

关 键 词：多重分形 HURST指数 语音情感 情感识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]