摘要
对于不平衡情感数据集,传统的模糊支持向量机原理上分类不灵敏,支持向量的隶属度值被给予不准确情况,提出一种对样本点赋值的设计方法,并将其用到语音情感识别。引入DEC算法,消除数据不平衡引起的分类超平面偏移的影响,计算从样本点到类中心超平面的距离,考虑样本周围的样本分布设计模糊隶属函数点。突出支持向量对分类超平面的贡献,有效降低噪声和孤立点的影响。实验结果表明,与传统的模糊支持向量机相比,对样本失衡率为4.89的TYUT2.0情感语音数据库的分类性能提高了5.95%,对不平衡率为14.28的CASIA中文情感语料库的分类性能提高了11.57%。
For the non-equilibrium emotion data set,the traditional fuzzy support vector machine is sensitive to classification and it lacks precision of the support vector.To solve the problems,the fuzzy membership function design method was proposed and applied to the speech emotion recognition.The unbalanced adjustment factor was introduced to eliminate the influence of the super-plane offset caused by the data imbalance.The distance between the sample point and the super-plane of the sample center was calculated,and the fuzzy membership function was designed by taking into account the density of the sample,highlighting the contribution of support vector to the classification hyperplane and effectively reducing the influence of noise and isolated point.The results show compared with the traditional fsvm,this way improves the classification performance by 5.95%on the TYUT2.0 emotion voice database with a sample imbalance of 4.89,and the classification of CASIA Chinese emotional corpus with the unbalanced rate of 14.28 was improved by 11.57%.
作者
张雪英
张波
陈桂军
ZHANG Xue-ying;ZHANG Bo;CHEN Gui-jun(College of Information Engineering,Taiyuan University of Technology,Jinzhong 030600,China)
出处
《计算机工程与设计》
北大核心
2018年第11期3544-3548,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61371193)
关键词
语音情感识别
模糊支持向量机
平衡调节因子
隶属度函数
样本密度
speech emotional recognition
fuzzy support vector machine(FSVM)
balance adjustment factor
membership function
sample density
作者简介
张雪英(1964),女,河北石家庄人,博士,教授,博士生导师,CCF会员,研究方向为语音信号处理;张波(1994),男,河北石家庄人,硕士研究生,研究方向为模糊支持向量机、语音识别;E-mail:zhangbo940101@126.com;陈桂军(1987),男,山西大同人,博士,讲师,研究方向为模式识别、支持向量机。
