Introducing New Feature Set based on Wavelets for Speech Emotion Classification

Feature extraction for Speech Emotion Recognition (SER) is a challenging task and there is no consensus among researchers on a single set of features which works best. Different speech features such as pitch, energy, formants, Mel-Scale Coefficients, Predictive Coding are used for classification but results are still not satisfactory enough. In this article, a new feature set is proposed which uses Discrete Wavelet Transform (DWT) to decompose the speech signal and computes dissimilarity with the neutral emotional state. The new feature set is used for emotion classification using three different classification techniques to establish that the feature set is giving better or competitive results compared to the contemporary features.