<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=UTF-8"> </head> <body text="#000000" bgcolor="#FFFFFF"> PhD Position <div style="text-align:center"> Learning and Assessment of Public Speaking in Virtual Reality </div> <div style="text-align:center"> Deadline for application: 8 May 2020 </div> <div style="text-align:center"> Laboratoire d’Informatique et des Systèmes (LIS) and Institut des Sciences du Mouvement (ISM) <del datetime="2018-03-09T13:30"></del> Aix-Marseille Université & CNRS</div> <div> </div> Keywords: Human-machine interaction, Behavioural data collection, Virtual Reality, Embodied Conversational Agent, Artificial intelligence, Machine Learning In the field of e-education and training, a growing interest has emerged around "training by simulation" based on virtual environments. Several studies have been conducted on devices to simulate social interaction with Embodied Conversational Agents (ECAs) to train one's own social skills ("Virtual Agents for Social Skills Training", Bruijnes et al., 2019; Ochs et al., 2019). Some research works have explored the use of a virtual audience composed of a set of ECAs to train individuals to speak in public (Chollet et al., 2015; Pertaub et al., 2002; North et al., 1998). One of the challenges in the development of virtual environments for the social skills training is to be able to automatically evaluate the user's performance and engagement in the task. In the context of public speaking, certain objective verbal and non-verbal cues may reflect the quality of public speaking; for example, the number of disfluencies in speech, the gestures and the gaze direction. User engagement, and more specifically in the field of virtual reality her/his sense of presence and co-presence, is also reflected through objective verbal and non-verbal cues (Ochs et al, 2019). The identification of behavioural cues of the quality of human-machine interaction is a major research issue. The objective of this thesis project is to explore objective behavioural and physiological cues of the quality of interaction in the context of the specific use case of public speaking in front of a virtual audience in a virtual reality (VR) environment. This is a major scientific challenge, based on work in both human movement science and computer science. The proposed methodology will be based on an interdisciplinary analysis of the objective cues of human-machine interaction in this situation, including language analysis (e.g. prosody, lexicon, syntax), analysis of gestures and postures (e.g. biomechanical measurement, facial expression) and physiological clues (e.g. measurement of heart rate, electro-dermal conductance). These interaction data will be analysed using machine learning methods to extract behavioural and physiological cues characteristic of the public speaking quality. The thesis will be organized around 4 main steps: 1/ Collection and semi-automatic annotation of a corpus of human-virtual auditory interaction in VR environment; 2/ Identification of social skills from the use case studied; 3/ Identification of the behavioural and physiological cues of social skills using machine learning techniques; and 4/ Development of an automatic evaluation system of social skills adapted to the use case. This interdisciplinary thesis will be carried out in collaboration with two research laboratories of Aix Marseille University: the <a moz-do-not-send="true" href="https://ism.univ-amu.fr/fr">ISM</a> and the <a moz-do-not-send="true" href="https://www.lis-lab.fr/">LIS</a>. The simulation technology platform developed at the <a moz-do-not-send="true" href="https://ism.univ-amu.fr/fr/crvm">Centre de Réalité Virtuelle Méditerranée</a> (CRVM) of the ISM for public speaking training in front of a virtual audience will be used. The evaluation will be carried out in collaboration with experts in the field (public speaking coach) to compare the output of the model with the expertise of the coaches. Required skills The candidate must have an initial training in Computer Science, or Movement Sciences or Cognitive Sciences. Skills in machine learning, experimental methods and statistical analysis are expected. Knowledge on Natural Language Processing (NLP), signal processing, motion capture, Python and R would be appreciated. Technical knowledge in programming (Java, C++, C Sharp) and Unity software would be a plus. A multidisciplinary background is a must. The thesis is fully funded by a 3-year doctoral contract (salary between 1400 and 1600 €/month) in the context of the Inter-Ed Interdisciplinary call of Aix Marseille University. The laboratories provide funding for professional travel, the candidate's training and participation in international conferences. French language is not required. Aix Marseille University (<a class="moz-txt-link-freetext" href="http://www.univ-amu.fr/en">http://www.univ-amu.fr/en</a>), the largest French University, is ideally located on the Mediterranean coast, and only 1h30 away from the Alps. The application files consists of the following documents: <ul> <li>A detailed curriculum,</li> <li>A motivation letter,</li> <li>A description of the academic background and copy of academic records and most recent diploma,</li> <li>A letter of recommendation from the internship supervisor or the director of the master's thesis.</li> <li>The master thesis 1 and/or 2. </li> </ul> The application files should be sent before the 8 may to Magalie Ochs: <a class="moz-txt-link-abbreviated" href="mailto:magalie.ochs@lis-lab.fr">magalie.ochs@lis-lab.fr</a> Rémy Casanova: <a class="moz-txt-link-abbreviated" href="mailto:remy.casanova@univ-amu.fr">remy.casanova@univ-amu.fr</a> For any question, you can contact us by mail. </body> </html>