基于类内超平面距离度量模糊支持向量机的语音情感识别
2018-08-21张波张雪英陈桂军孙颖
张波 张雪英 陈桂军 孙颖
摘 要: 在智能人?机交互系统中,语音情感识别是目前的研究热点之一,支持向量机方法被广泛用于语音情感识别。然而,支持向量机方法存在噪声和野值敏感问题,往往难以进行精确识别。为了解决该问题,通过对隶属度函数进行深入研究,设计一种新的基于样本到类内超平面距离的隶属度函数,并基于该隶属度函数优化了模糊支持向量机分类超平面,从而提高了支持向量机的抗噪性和泛化能力。在多种情感语音库上进行实验仿真测试,结果表明,所提出的方法能够有效利用样本间的紧密度、边界样本点和过样本类中心的超平面来构造最优超平面,从而提高语音情感识别的准确率。
关键词: 语音情感识别; 模糊支持向量机; 隶属度函数; 孤立点; 类内超平面; 精确识别
中图分类号: TN912.34?34; TP181 文献标识码: A 文章编号: 1004?373X(2018)16?0163?05
Abstract: In the intelligent human?machine interaction system, speech emotion recognition is one of the current research hotspots. The support vector machine method is widely used in speech emotion recognition, but it has problems of noises and wild value sensitiveness, resulting in difficulty of accurate identification. Therefore, a novel membership function based on the distance from samples to the intra?class hyperplane is designed by means of the in?depth study of the membership function, based on which the classification hyperplane is optimized by using the fuzzy support vector machine, so as to improve the anti?noise and generalization capabilities of the support vector machine. An experiment and simulation test were carried out by using various emotion speech libraries. The experimental results show that the proposed method can effectively utilize the sample compactness, boundary sample points, and the hyperplane passing through the center of the sample class to construct the optimal hyperplane, which can improve the accuracy of speech emotion recognition.
Keywords: speech emotion recognition; fuzzy support vector machine; membership function; isolated point; intra?class hyperplane; accurate recognition
0 引 言
随着人机交互技术的发展,情感识别技术已经日益成为科研人员研究的焦点,比如智能人机交互[1?2]、疾病诊断和测谎仪等。
语音信号的情感识别方法有很多,常用的情感分析方法有混合高斯模型法(GMM)、隐马尔科夫模型法(HMM)[3?4]、人工神经网络方法(ANN)[5]以及支持向量机(SVM)等。其中SVM情感分析方法在解决非线性、小样本以及高维模式识别问题中有着良好的分类效果,但是从本质上来说SVM是一种不适当问题的正则化理论和非线性规划计算方法,在情感混淆程度较大的情况下,往往难以进行精确识别。
文献[6]将隶属度的概念引入到SVM分类中,提出了模糊支持向量机。……
