Emotion Recognition in Speech Using Machine Learning Techniques

Abstract

This paper contains an analysis of the performance of various machine learning models in detecting and labelling emotion in speech on existing and novel emotional speech corpora, in conjunction with various feature sets. The emotions considered are happiness, sadness, neutrality, anger, fear, and sarcasm. This paper introduces a novel dataset composed of 900 clips in various Indian languages from a wide range of sources ranging from politicians' speeches to movies and comedy shows. Given the diversity of Indian languages, this dataset is a step towards building more language-agnostic models that could be useful in multilingual countries. This paper also includes sarcasm detection, a challenging problem to combat in the field of emotion recognition in speech. Further, this paper aims to find the best model-feature combination for emotion recognition across all the considered datasets.