Next Department 4 seminar will take place on Thursday, 1st December at 11am in room B011-B013.

Jean-Luc Rouas (CNRS – LaBRI): An overview of speech processing research at LaBRI

Vincent Martin (Université de Bordeaux – LaBRI / SANPSY): Non-invasive monitoring of sleepiness from speech
Excessive sleepiness is both a personal and public health problem, responsible for almost a third of motorway accidents in France. However, the imbalance between the prevalence of excessive sleepiness and the number of medical specialists, and the need to monitor patients between their consultations and in their usual living conditions, has led sleep medicine to invest in the Ecological Momentary Assessment. Voice and speech are thus very promising tools for non-invasively monitoring the evolution of patients’ symptoms in their daily lives. We propose in this presentation a brief summary of the work that we have carried out at the Clinique du Sommeil de Bordeaux, under very controlled conditions, on hypersomniac patients that we have recorded during reference medical tests (Iterative Sleep Latency Test). Based on read speech recordings, we designed and extracted new descriptors automatically extracted from the voice and speech of these patients, which we related to their physiological (EEG) and subjective (medical questionnaire) sleepiness. Finally, we present our research perspectives, notably around the opening of this work towards more ecological conditions and the joint estimation of the different symptoms expressed in the voice, thanks to symptom networks.

Florian Pecune (Université de Bordeaux – SANPSY): Socially-aware conversational agents
People pursue multiple conversational goals when they interact with each other. The task goals are related to the informational content of the conversation and define its purpose. On the other hand, social goals are related to engagement and relationship management. A similar dichotomy in terms of conversational goals exists in human-computer interactions. While personal assistants such as Siri, Cortana, Alexa, etc., focus on the task-oriented goals of the conversation by providing answers to their users questions, chatbots such as Eliza or XiaoIce focus on the interpersonal aspect of the conversation and try to engage their users as long as they can. In this talk, I will present my work in the area of socially-aware conversational agents. More specifically, I will focus on how our socially-aware conversational agents (1) detect user’s social behaviors in conversation, (2) reason about how to respond to the intentions behind those particular behaviors, and (3) generate appropriate social responses – all that while carrying out their task duties at the same time. I will also present Kanopee, the conversational agent developped by our team at the University of Bordeaux to help people complaining about their sleep.