Performance Analysis of MFCC Features On Emotion Recognition from Speech

Abstract

This work presents MFCC-based emotion recognition from speech. For this purpose, features of labeled speech signals are extracted and the classifier is trained. Then, test data is classified using its features and the classification performance is measured. In this work, MFCC (Mel-frequency Cepstrum Coefficients) features are extracted for training and recognition. In addition to investigation of success rates for different emotion classes, comparison of results to other results obtained with additional features are also analyzed.

Keywords: emotion recognition from speech, performance analysis on emotion recognition, emotion classification.