Voice Response Based Emotion Intensity Classification for Assistive Robots

Abstract

A service robot should be able to recognize the user's emotional status in order to maintain the human friendliness of its responses. Understanding emotions is an immense challenge for a robot since the emotion are difficult to represent in a quantified domain. Therefore, this paper proposes a method to calculate the intensity level of emotion through machine-learning models and classify user emotions into four categories. The same objective was achieved with four machine learning models, and their performances were analyzed across a variety of parametric evaluations. Finally, the best algorithm that performs well across all the parameters is chosen. The proposed system performs significantly better on test data, with an accuracy of 99.47%. In addition, an experimentation-based study was also held to compare the proposed system's predictions with human participants. The results of the experimental studies reveal that the proposed system's predictions have significant alignment with the data extracted from the responses from the human participants.