Speech emotion recognition using SVM classifier with suprasegmental MFCC features

Abstract

This study explores speech emotion recognition (SER) using mel-frequency cepstral coefficients (MFCCs) and Support Vector Machines (SVMs) classifier on the RAVDESS dataset. We proposed a model which uses 80-component suprasegmental MFCC feature vector as an input downstream by SVM classifier. To evaluate the quality of the model, unweighted average recall (UAR) was used. We evaluate different kernel functions for SVM (such as linear, polynomial and radial basis)and different frame size for MFCC extraction (from 20 to 170 ms). Experimental results demonstrate promising accuracy(UAR = 48%), showcasing the potential of this approach for applications like voice assistants, virtual agents, and mental health diagnostics.