Comparison between five classification techniques for classifying emotions in human speech

Abstract

This paper presents an algorithm for recognition of emotions in speech by extracting features such as formants, Perceptual Linear Prediction coefficients, Mel-Frequency Cepstral Coefficients, Bark Frequency Cepstral Coefficients, energy, pitch and standard deviation. The classifiers implemented are K-Nearest Neighbors (KNN), Linear Support Vector Machine (SVM), Quadratic SVM, Bagged Tree Ensemble and Quadratic discriminant. The paper presents a comparative study on the different classification techniques that can be used to distinguish between various emotions present in human speech. A comparison in terms of testing accuracy obtained using these classifiers has been performed in this paper on a database created for 4 emotions viz. anger, joy, sorrow and neutral in Marathi language.