Towards an Intelligent Emotional Detection in an E-Learning Environment

Abstract

Several research in psycho-pedagogy showed that the relevance of the training implies as many intellectual aspects as socio-emotional and the emotional state of the learner influence directly his performance in a positive or negative way. For that, the e-learning systems must take into account the emotional state of learners in order to favor their training. However, it is important to underline the implementation problem which is posed and summarized in the detection and the interpretation of the emotions which are not directly observable by the machine and which are generally expressed by a whole of behaviours whose indicators are either the words used, or the voice ton, the gestures and body attitudes or facial expressions. In this paper, we propose emotionally intelligent system architecture dedicated to the learning activities. We focus ourselves more particularly on the process of emotional recognition, ensured by agent EMOTIO, from a bimodal analysis of the speech and text used by a speaker’s learner, in a training session, in order to improve the recognition precision (precision in acoustical analysis is 63.44 % but precision in bimodal analysis is 71.2%). This analysis is based on the indices extraction on two linguistic levels: prosodic and lexical.