PHD position at LIMSI-CNRS

Context:

The most advanced research in robotics aims at implementing robots interacting in a more and more natural way with humans. To give a robot the means of recognizing the person interacting with it and of perceiving his (her) emotional states should allow a more natural interaction. Indeed, coupled with a personality's model and with an analysis of the interactive situation, the robot can express sensible feelings with regard to the situation. The robot will behave differently with a child, an adult or an old person, differently if the situation is serious or playful, differently if its strategy is to be very empathic or not.

Can a robot detect and interpret an emotion? The concepts and the methods of the artificial intelligence, the linguistics, the psychology and the cognitive sciences begin to bring elements of answer and allow the development of computer systems with an emotional intelligence. This subject is an emerging domain which offers numerous scientific challenges. This PhD is financed in the context of the French project ROMEO headed by Aldebaran Robotics (project accredited with the Digital Cap Label). We will use at LIMSI the open, programmable and interactive platform NAO already used for research purposes, which is an ideal robot for emotional and affective computing.

Subject:

The subject concerns the characterization of the speaker which interacts with the robot, that is, the identification and the focus on the speaker (s) and the detection of its (their) feelings from non verbal cues, mainly audio cues. The exploration of multimodal indications like the facial expressions for both tasks can also be possible.

The speaker recognition task addresses the identification of one or several speakers: vocal signature (who speaks), the focus on a specific speaker (voice of his owner), the qualification of the strangers (guess the age and the sex of the strangers). Two major challenges will be the use of the robot in a noisy acoustic environment, and the interactions between the identification of the speaker and his emotional states.

The emotion detection task in the voice addresses the detection of various emotional states as the moods and the feelings. The challenges are situated at the level of the characterization of the affective states in the collected data, the identification of the audio cues characterizing the emotional states and finally in the implementation of real-time emotion detection systems.

To apply, include the following information: - C.V, - the names and contact email/phone of one or two academic referees who would willing to write letters of recommendation, - and a cover letter describing your research interests

L. Devillers, L. Vidrascu, L. Lamel, (2005) “Emotion detection in real-life spoken dialogs recorded in call center”, Journal of Neural Networks, numéro spécial “Emotion and Brain”, volume 18, numéro 4, 407-422, mai 2005.

F Bimbot et al. (2004), a Tutorial on Text-Independent Speaker Verification, journal Eurasip, 2004