October 2024

Date: 21th of october 2024 at 2PM Place: Thesis room, at the Hannah Arendt campus of Avignon Université. The videoconference link is the following: https://bbb.univ-avignon.fr/rooms/vtj-xje-xex-gyw/join .  The jury will be composed of: Dr Aurélie Clodic, LAAS-CNRS,  RapporteurePr Julien Pinquier, Université de Toulouse, IRIT, RapporteurPr Laurence Devillers, Sorbonne Université, LISN-CNRS, ExaminatricePr Olivier Alata, Université Jean Monnet, Laboratoire Hubert Curien, ExaminateurPr Fabrice Lefèvre, Avignon Université, LIA, Directeur de thèseDr Bassam Jabaian, Avignon Université, LIA, Co-encadrant Title: Proactive multimodal human-robot interaction in a hospital In this thesis, we focus on creating a proactive multimodal system for the social robot Pepper, designed for a hospital waiting room. To achieve this, we developed a cognitive human-robot interaction architecture, based on a continuous loop of perceptions, representation, and decision-making. The flow of perceptions is divided into two steps: first, retrieving data from the robot’s sensors, and then enriching it through refining modules. A speaker diarization refining module, based on a Bayesian model of fusion of audio and visual perceptions through spatial coincidence, was integrated. To enable proactive action, we designed a model analyzing the users’ availability for interaction in a waiting room. The refined perceptions are then organized and aligned to create a constantly updated representation of Plus d'infos

Date:  Jeudi 24 octobre à 15h Lieu: salle des thèses sur le campus Hannah Arendt. Vous pouvez également y assister à distance si vous le souhaitez grâce au lien suivant : https://v-au.univ-avignon.fr/live/bbb-soutenance-these-l-druart-24-octobre-2024/. Title : Towards Contextual and Structured Spoken Task-Oriented Dialogue Understanding Abstract : Accurately understanding users’ requests is key to provide smooth interactions with spoken Task-Oriented Dialogue (TOD) systems. Traditionally such systems adopt cascade approaches which combine an Automatic Speech Recognition (ASR) component with a Natural Language Understanding (NLU) one. Yet, those systems still have trouble to accurately map complex user’s request with their internal representation. Recent work highlights potential directions to improve those systems. On the one hand, end-to-end approaches have successfully enhanced Spoken Language Understanding (SLU) system’s performance. Indeed, they provide more robust and accurate predictions by leveraging joint optimization and paralinguistic information. On the other hand, textual datasets propose fine-grained semantic representations. Such representations seem more adequate to represent user’s complex requests. This thesis explores both directions towards contextual and structured spoken task-oriented dialogue understanding. We first conduct a preliminary study dedicated to getting the grips of SLU in the context of TOD. We designed a cascade approach to perform spoken Dialogue State Tracking (DST) on MultiWOZ. Our approach ranked first in Plus d'infos