Abstract: In order to improve the accuracy of emotion recognition, a novel emotion recognition method（MF-KCCA） based on multi-features fused by kernel canonical correlation analysis to solve the defects of single feature and traditional features fusion method is proposed. The speech prosody and fractional Fourier domain features are extracted, and then two kinds of features are fused by KCCA to reduce the dimension of feature vector. Emotion is recognized by the nearest neighbor classifier. The simulation experiments are carried out on Canadian Ryerson University database, and the results show that the proposed method can effectively improve the emotion recognition rate.

Key words: emotion recognition, feature fusion, emotion features, prosodic features, kernel canonical correlation analysis

摘要： 为了提高情感识别的正确率，针对单模情感特征及传统特征融合方法识别低的缺陷，提出了一种核典型相关分析算法（KCCA）的多特征（multi-features）融合情感识别方法（MF-KCCA）。分别提取语音韵律特征和分数阶傅里叶域表情特征，利用两种特征互补性，采用KCCA将它们进行融合，降低特征向量的维数，利用最近邻分类器进行情感分类和识别。采用加拿大瑞尔森大学数据库进行仿真实验，结果表明，MF-KCCA有效提高了语音情感的识别率。

关键词: 情感识别, 特征融合, 表情特征, 韵律特征, 核典型相关分析