Title: Decoding speech non-invasively

Speaker: Oiwi Parker-Jones (Faculty Department of Engineering Science, University of Oxford; Member Information Engineering; Principal Investigator Oxford Robotics Institute; Head Parker Jones Neural Processing Lab (PNPL); Hugh Price Fellow in Computer Science, Jesus College, University of Oxford; Honorary Fellow Nuffield Department of Clinical Neuroscience)

Location: In Person & Online via Zoom

Abstract: There are moments in science with new frontiers open and what was once science fiction becomes real. In just the past few years, it became possible to transcribe neural representations of speech, i.e. brain-to-text (B2T) decoding, using surgically implanted electrodes. This is a momentous achievement with profound implications for paralysed individuals; but I would still advise caution before volunteering for a craniotomy. While methods like M/EEG have much lower SNR than direct brain recordings, they are dramatically less risky for users. They also significantly easier to scale non-invasive data and then leverage deep learning. In this talk, I will discuss the current state of the field and present work from my group (PNPL) which is driving advances in non-invasive neural speech decoding. There are three parts to our strategy: scaling up data, developing neural decoding methods, and investigating the neurobiology of inner speech.

Website / BlueSky: https://pnpl.robots.ox.ac.uk/