基于KPCA和模糊核Fisher判别的语音情感识别

针对语音情感识别率不高和实时性差的问题,提出一种基于KPCA核主成分空间的模糊KFD算法,应用于语音情感识别。首先采用KPCA对语音情感特征向量降维去噪,根据转换矩阵得到核主成分空间,然后在该特征空间利用模糊C均值聚类计算语音特征向量的隶属度,进而对LDA算法中的类间离散度和类内离散度重新定义,生成模糊KFD分类器进行语音情感识别。仿真实验结果表明,提出的方法相比于传统SVM和核Fisher判别算法具有较高的识别率和良好的抗噪性能,是一种行之有效的语音情感识别新方法。

Speech Emotion Recognition Based on KPCA and Fuzzy Fisher Discrimination

A kind of novel fuzzy Kernel Fisher Discrimination(KFD) algorithm based on principal component space of Kernel Principle Component Analysis(KPCA) is proposed to improve speech emotion discrimination rate and real-time processing and is used in speech emotion discrimination.KPCA was firstly used to reduce dimensions of and remove noises for speech feature vector,kernel principal component space was obtained according to transformation matrix,then in this feature space,fuzzy C-mean cluster was utilized to compute the menbership of each speech feature vector,and furthermore,inter-calss divergence factor and in-calss divergence factor in LDAalgorithm were redefined to form fuzzy KFD calssifier to conduct speech emotion disctimition.Simulation experiment results show that the proposed method in this paper hais higher recognition rate and better anti-noise performance than traditional SVM and kerenl Fisher discriminatin algorithm and is a kind of novel effective method for speech emotion discrimination.