Fifth International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2007)

Florence, Italy
December 13-15, 2007

Real-Life Emotions Detection on Human-Human Medical Call Center Interactions

L. Devillers, L. Vidrascu

Department of Human-Machine Communication, LIMSI-CNRS, Orsay, France

Our aim is to study the vocal expression of emotion in real-life spoken interactions in order to build emotion detection system. We make use of a corpus of naturallyoccurring dialogs recorded in a real-life emergency medical call center. The context of emergency gives a large palette of complex and mixed emotions. About 30% of the utterances are annotated with non-neutral emotion labels on this medical corpus. The complexity of the emotion recognition task increases the higher the number of classes and the finest and closest these classes are. Finding relevant features of various types such as speech disfluencies or affect bursts becomes essential in order to improve the detection performances. Our experiments focus on a task of discriminating 2 to 5 emotions, Fear, Anger, Sadness, Neutral and Relief.
Index Terms. Emotions, real-life spoken interactions, detection system, medical call center

Full Paper (reprinted with permission from Firenze University Press)

Bibliographic reference.  Devillers, L. / Vidrascu, L. (2007): "Real-life emotions detection on human-human medical call center interactions", In MAVEBA-2007, 143-146.