Emotion Recognition from Speech for an Interactive Robot Agent

Abstract

Speech is one of the fundamental approaches for human to human interaction. Given this, it should be the main approach for robot human interaction as well. Towards this, the research presented here focuses on emotion recognition from human speech to aid the interaction between humans and robots. There are various steps involved in developing emotion recognition system for an interactive robot agent. The first step is to choose a suitable dataset that is Berlin database for training and testing the models developed. The second important step is extraction and choice of suitable features related to emotions. The third step is to make an appropriate classification scheme. The performance of each classifier is analyzed and acomparison among multiple frameworks of emotion recognition is made. In response to the findings in these preliminary studies, a prototype application was developed to allow the recognition of emotions from speech in real-time for future use on an interactive robot. On a preliminary test set, the application achieved performance levels between 81% and 92%. The approach offers the integration of speech collection hardware, emotion recognition software, mobile devices and robotic systems to aid assist human-robot interaction.