MFCC and Machine Learning Based Speech Emotion Recognition Over TESS and IEMOCAP Datasets

Abstract

Emotions in speech provide a lot of information about the speaker’s emotional state. This paper presents a classification of emotions using a support vector machine (SVM) with Mel Frequency Cepstrum Coefficient (MFCC) features extracted from the voice signal. We have considered the following five emotions, namely anger, happy, neutral, pleasant surprise and sadness, for classification purposes. The proposed methodology, including SVM-Gaussian and SVM-Quadratic, is tested for its performance on the Toronto Emotion Speech Set (TESS) and Interactive Emotional Dyadic Motion Capture (IEMOCAP) datasets. Our proposed methodology achieved 97% accuracy with TESS and 86% with IEMOCAP datasets, respectively.