Speech Emotion Recognition: Models Implementation & Evaluation

Abstract

Speech emotion recognition (SER) is a new emergent field of research that has several possible applications in both human-computer and human-human interaction systems. The body of work on emotion detection from speech signal is relatively limited. Nowadays, researchers are yet augmenting on what features effects the identification of emotion in speech. There is a significant ambiguity as to the finest algorithm for emotion's classification, and which emotions to class together as well. In this work, we seek to address these matters. We use Support Vector Machines (SVMs) and Artificial Neuron Network (ANN) to classify opposite emotions. There is a variability of temporal and spectral characteristics that can be extracted from human speech. We focus only on Mel Frequency Cepstral Coefficients (MFCCs) as inputs to the classification algorithms. The classification reports obtained from the conducted experiments allow us to say that, for the given parameters, the ANN model was better detecting the speech carried emotional information than the SVM.