基于特征降维及参数优化的语音情感识别  

Speech emotion recognition based on feature dimension reduction and parameter optimization

作　　者：俞颖 黄风华 刘永芬[3] YU Ying;HUANG Fenghua;LIU Yongfen(Spatial Data Mining and Application Research Center of Fujian Province,Yango University;Artificial Intelligence College,Yango University;Jinshan College,Fujian Agriculture and Forestry University:Fuzhou 350001,China)

机构地区：[1]阳光学院空间数据挖掘与应用福建省高校工程研究中心 [2]阳光学院人工智能学院 [3]福建农林大学金山学院,福建福州350001

出　　处：《延边大学学报（自然科学版）》2020年第1期49-54,共6页Journal of Yanbian University（Natural Science Edition）

基　　金：福建省教育厅中青年教师科研项目(JAT190977);福建省自然科学基金资助项目(2019J01088)

摘　　要：针对传统BP神经网络在语音情感识别过程中存在的计算量偏大和容易陷入局部最优解的问题,提出了一种基于特征降维及参数优化的情感识别改进方法.首先提取情感语料库的高维度联合特征,利用快速主成份分析法(Fast_PAC)进行特征降维以达到降低问题复杂性的目的;然后引入遗传算法对BP神经网络进行参数优化以避免限入局部最优问题;最后构建语音情感识别分类器,并利用CASIA汉语语料库及柏林德语语料库进行情感识别验证.验证结果表明,与传统的支持向量机(SVM)方法、传统的主成份分析法(PCA算法)结合SVM模型识别方法相比,本文方法能有效地降低语音情感的特征维度,且具有运算量少和识别精度高的优点.The traditional BP neural network has been existing some burning questions in the process of speech emotion recognition,especially the high computational and local optimum trending.Against these shortcomings,we present a novel method of emotion recognition based on feature dimension reduction and parameter optimization.The recognition method is divided into three stages.In the first stage,it extracts the high-dimensional joint features of the speech emotion database.This is,in fact,aimingto reduce the complexity of the problem which is carried out by the fast principal component analysis(Fast_PAC)method.In the second stage,genetic algorithm is used to optimize the parameters of BP neural network to avoid the local optimum problem.Finally,we construct a speech emotion recognition classifier,and take the experiments on the CASIA Chinese corpus and Berlin German corpus for emotion recognition verification.The experiments show that the proposed method can effectively reduce the feature dimension of speech emotion comparing with other competitive methods,such as the traditional support vector machine(SVM)method and the traditional PCA combined with SVM model recognition method.Furthermore,it demonstrates the advantages of less computation and higher recognition accuracy.

关 键 词：快速主成份分析法 遗传算法 BP神经网络 语音情感识别 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]