论文部分内容阅读
针对说话人无关的语音情感识别,提出一个分层语音情感识别模型,由粗到细识别悲伤、愤怒、惊奇、恐惧、喜悦和厌恶6种情感.每层采用Fisher比率从288个备选特征中选择适合该层分类的特征,同时将Fisher比率作为输入参数训练该层的支持向量机分类器.基于北京航空航天大学情感语音数据库和德国柏林情感语音数据库,设计4组对比实验,实验结果表明,Fisher准则在两两分类特征选择上优于PCA,SVM在说话人无关的语音情感识别推广方面优于人工神经网络(ANN).在两个数据库的基础上得到类似结果,说明文中分类模型具有一定的跨文化适应性.
For speaker-independent speech emotion recognition, a hierarchical speech emotion recognition model is proposed, which identifies the six emotions of sadness, anger, surprise, fear, joy and disgust from coarse to fine. Each layer adopts Fisher’s ratio from 288 candidate features Choose the suitable features of this layer classification and train the Fisher SVM classifier as the input parameter.Based on the sentiment database of Beijing University of Aeronautics and Astronautics and the emotional database of Berlin in Germany, four groups of contrast experiments are designed, and the experimental results show that, The Fisher criterion is superior to PCA in the feature selection of pairwise classification and the SVM is superior to artificial neural network (ANN) in extension of speech emotion recognition unrelated to speaker.The similar results are obtained on the basis of two databases, which shows that the classification model has a certain Cross-cultural adaptability.