HMM modeling of user mood through recognition of vocal emotions

Abstract

This paper aims at defining a real-time probabilistic model for user's mood in its dialect with a software agent, which has a long-term goal of counseling the user in the domain of "coping with exam pressure". We propose a new approach based on Hidden Markov Models (HMMs) to describe the differences in the sequence of emotions expressed due to different moods experienced by users. During real time operation, each user move is passed on to a vocal affect recognizer. The decisions from the recognizer about the kind of emotion expressed are then mapped into code-words to generate a sequence of discrete symbols for HMM models of each mood. We train and test the system using corpora of the temporal sequences of tagged emotional utterances by six male and six female adult Indians in English and Hindi language. Our system achieved an average f-measure rating for all moods of approximately 78.33%. © 2013 ICST Institute for Computer Science, Social Informatics and Telecommunications Engineering.