91社区

Unpacking the Speech Chain: A window of scientific and technological opportunities

Satrajit Ghosh, MIT
Tuesday October 18, 12-1pm
Zoom Link:聽

Abstract:聽Speaking is one of the most complex tasks carried out by humans and we do it, mostly, effortlessly. We use it to communicate our thoughts, feelings, and intent. In our group we have targeted this model system as a window into the human brain to understand basic brain and behavioral mechanisms, and as a rich signal for developing new technologies and applications. Lately, there has been a significant increase in the attention to and relevance of this seemingly simple timeseries thanks in large part due to pervasive growth of smartphones, voice assistants, and machine learning technology. In this talk, I will present the connectedness and current knowledge of spoken communication in relation to ongoing projects on stuttering, on identity, and on mental health. Each of these projects started with a simple question that drew us in through different angles to understand the system better. Mumble melody, the project on stuttering explores mechanisms to induce greater fluency in people who stutter. A question of quantifying how similar two people sound led us to a project on identity and privacy. Evidence of speech signals related to neuropsychiatry and neurology has spurred intersecting speech and linguistics into mental health and wellbeing. Each of these projects have made us question existing knowledge and models and has opened doors for new science, technology, and education. I will end by discussing the goals of a new consortial project recently launched by the US National Institutes of Heath to study and evaluate voice as a biomarker across health. This project will focus on generating a rich, diverse, standardized, and ethically considered dataset that accelerates the development of reduced-bias machine learning algorithms and technologies that can be used scalably, reliably, and remotely, and address individuals across social determinants of health.