摘要
针对语音情感识别率不高的问题,提出一种基于MVR隶属度的多级FSVM算法。在采用PCA对输入特征向量进行约简的同时得到话者的主分量空间,在此空间对注册说话人进行初次筛选,并根据每位话者的特征向量在主分量空间上的映射方差比来计算该特征向量属于该类的模糊隶属度,最终使用MVR-FSVM进行语音情感识别。仿真实验结果表明,多级MVR-FSVM具有良好的分类性能和抗噪性能。
A novel hierarchical FSVM speech emotion recognition approach based on MVR member- ship was proposed in this paper. Firstly, PCA was utilized to reduce the dimension of registered speakers＇ feature vectors, and simultaneously principal component space was obtained based on the transform matrix. Then PCS-PCA classifier was proposed to select the possible R target speakers fleet- ly. After that, the membership of FSVM was computed using mapping variance ratio of the speaker＇ s input vector X. Finally, the recognition result was achieved by MVR-FSVM. The experiment results validate availability and feasibility of our method.
出处
《重庆理工大学学报（自然科学）》
CAS
2012年第10期66-71,共6页
Journal of Chongqing University of Technology：Natural Science
基金
甘肃省教育厅资助项目(1113-01)
关键词
语音情感识别
主成分分析
模糊支持向量机
speech emotion recognition
principal component analysis
fuzzy support vector machine
