Speech emotion recognition: Features and classification models

摘要: To solve the speaker independent emotion recognition problem, a three-level speech model is proposed to classify six emotions, including sadness, anger, surprise, fear, happiness and disgust from coarse fine. For each level, appropriate features are selected 288 candidates by using Fisher rate which also regarded as input parameter for Support Vector Machine (SVM). In order evaluate system, principal component analysis (PCA) dimension reduction artificial neural network (ANN) classification adopted design four comparative experiments, Fisher+SVM, PCA+SVM, Fisher+ANN, PCA+ANN. The experimental results proved that better than PCA reduction, SVM more expansible ANN recognition. average rates level 86.5%, 68.5% 50.2% respectively.