Speech emotion classification using SVM and MLP on prosodic and voice quality features

Abstract

In this paper comparisons emotion classification between Support Vector Machine (SVM) and Multi Layer Perception (MLP) Neural Network using prosodic and voice quality features extracted from Berlin Emotional Database are reported. The features were extracted using PRAAT tools while WEKA tool was used for classification. Different parameters set up for both SVM and MLP were implemented in getting the optimized emotion classification. The results show that MLP overcomes SYM in overall emotion classification. Nevertheless, the training for SYM was much faster compared to MLP. The overall recognition rate was (76.82%) for SYM and (78.69%) for MLP. Sadness was the highest emotion recognized by MLP with recognition rate of (89.0%) while anger was the highest emotion recognized by SYM with recognition rate of (87.4%). The most confusing emotion using MLP classification were happiness and fear while for SYM, the most confusing emotions were disgust and fear.