Speech Emotion Recognition Based on Kernel Principal Component Analysis and Optimized Support Vector Machine

Speech emotion recognition is a hot topic nowadays. In order to improve recognition rates, we propose a new solution based on kernel principal component analysis (KPCA) and optimized SVM. KPCA reduces features dimension, while the gravity search algorithm (GSA) is used to optimize support vector machine (SVM) classification performances. We show that KPCA with Gaussian kernel function applied to the Berlin emotional speech database, KPCA outperforms the classical principal. On the other hand, we present experimental results showing that the GSA-SVM model achieves higher recognition rates and runs faster than two classical optimized SVM namely the SVM-particle swarm and the SVM-genetic algorithm. Combined together the KPCA-GSA/SVM achieves good recognition rates.