Affective Computing for Social Companion Robots Using Fine-grained Speech Emotion Recognition

Abstract

The increasing demand and diverse applications for social companion robots necessitate the development of more engaging and meaningful human-robot interactions and hence affective computing or emotion Al. In this paper, we propose a fine-grained speech emotion recognition using a state-of-the-art Deep Convolutional Neural Network trained on three-channel representations of speech signals to classify each emotion and also their intensity level. Experimental results on a publicly available dataset with intensity level (RAVEDESS) show that our method can effectively predict the users emotion and their intensity with 95.85±1.38% accuracy, a promising results towards empowering companion robots to be more affective and potentially be helpful in emotion regulations of their users.