AUDITORY AFFECT RECOGNITION

Abstract

The study of emotion recognition to lay groundwork for this work took a kick start by hovering through the so far conducted research driven by other researchers. According to a saying, one must look at the weather when they step out and observe people’s faces when they step in, recognition of correct facial expressions (FEx) is a way to infer someone’s inner emotions and intentions. However, assessing emotions just from FEx, unquestioningly could be a naïve lead as people may suppress their true emotions by articulating false FEx. Since, humans don’t always wear their hearts on their sleeves, we can shift the source from FEx to voice or speech to get a better exactness of the true emotion. An emotion recognition system using speech (SER) makes a significant contribution in the field of computer vision in terms of assessing human behaviors using machines. In SERs, both accuracy and interpretability form equal value. In this paper, we aim at solving this issue by combining interpretability of Fuzzy rule based systems with accurateness of estimations done on the basis of Kernel Densities. The RAVDESS dataset has been used to analyze the performance by considering variants of possible combinations of the dataset entries.