Analysis and Evaluation of Discriminant Analysis Techniques for Multiclass Classification of Human Vocal Emotions

Many of the classification problems in human computer interaction applications involve multi class classification. Support Vector Machines excel at binary classification problems and cannot be easily extended to multi class classification. The use of Discriminant analysis how ever is not experimented widely in the area of Speech emotion recognition. In this paper Linear Discriminant Analysis and Regularized Discriminant Analysis are implemented over Berlin and Spanish emotional speech databases. Prosody and spectral features are extracted from the speech database and are applied individually and also with feature fusion. Based on the results obtained, LDA classification performance is poor than RDA due to the singularity problem. The results are analysed using ROC Curves.