Abstract

In this talk, I will outline the unique application of speech technologies to brain signals collected from the cortical surface of clinical human subjects speaking isolated utterances, which are to be decoded in real time. I will introduce the techniques and challenges involved in dealing with this type of data. I will cover approaches I’ve applied within our lab, from simple, direct template matching of a small set of keywords to flexible, frame-level voice activity detection (VAD) and speech synthesis via deep neural networks (DNNs) from brain signals. I will then outline our recently begun clinical trial, including its unique challenges, some initial successful applications from our group and intended future directions.