摘要
在多种情感的语音情感识别中,由于部分情感状态容易混淆,导致语音情感识别的总体识别率降低;同时,对于不同的训练集,SVM参数惩罚因子和核函数参数对识别结果也存在一定影响。为了有效提高语音情感识别系统的识别率,在传统支持向量机(SVM)的基础上,提出了一种基于参数寻优决策树SVM的语音情感识别方法。该方法首先通过计算情感混淆度构建决策树SVM框架,然后采用遗传算法对决策树SVM中每个SVM的惩罚因子和核函数参数进行寻优,最后将参数优化后的决策树SVM模型应用于语音情感识别。在中文情感语音库的实验结果表明,与传统基于SVM分类方法的语音情感识别进行对比,该方法可将六种情感的平均识别率提高6.5%。
In the multi-emotion speech emotion recognition,the partial recognition rate is reduced because of the confusion of some emotional states.At the same time,the penalty factors and kernel function parameter also have some influence on the recognition results for different training sets.On the basis of traditional support vector machine（ SVM）,we propose a decision tree SVMspeech emotion recognition algorithm based on parameter optimization for improving the accuracy of speech emotion recognition.In this algorithm,the decision tree SVMframework is firstly established by calculating the confusion degree of emotion.Then the genetic algorithm is used to optimize the penalty factor and kernel function parameters of each SVMin the decision tree SVM.Finally,the decision tree SVMmodel with optimized parameters is applied to speech emotion recognition.The experiment on the Chinese emotion speech database shows that the proposed method can improve the average recognition rate of 6 emotions by 6. 5%,compared with speech recognition based on traditional SVMclassification algorithm.
作者
王富
孙林慧
苏敏
赵城
WANG Fu;SUN Lin-hui;SU Min;ZHAO Cheng(Key Lab of Broadband Wireless Communication and Sensor Network Technology of Ministry of Education,School of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)
出处
《计算机技术与发展》
2018年第7期63-67,共5页
Computer Technology and Development
基金
国家自然科学基金(61501251
61571233)
江苏省自然科学基金(BK20140891)
南京邮电大学校科研基金(NY214038)
关键词
语音情感识别
情感混淆度
决策树SVM
遗传算法
参数寻优
speech emotion recognition
emotional confusion
decision tree SVM
genetic algorithm
parameter optimization
