Speech-Based Emotion Recognition Using Sequence Discriminant Support Vector Machines

Abstract

Automatic Emotion Recognition (AER) is an interesting and recent research topic in the Human- Computer Interaction (HCI) field. In this paper a speaker- independent Automatic Human Emotion Recognition system is presented which is able to classify six discrete emotional states: happiness, sadness, anger, surprise, fear, and disgust. A set of novel and robust acoustic features are presented which are proved to yield a very good result. Least Square- Support Vector Machines (LS-SVMs) are proposed as a very powerful classifier with many advantages over other popular classifiers. In order to be able to discriminate between the whole sequences rather than frames, the use of Fisher kernels which make use of the information in the underlying generative models is suggested. Fuzzy-pairwise method is implemented to extend the binary SVMs to our multi-class problem. The overall classification rate of 97.65% is achieved.