This looks super neat. We’ve been looking into detect sub vocal communication for a while. If this really does work then it’s a far easier technique than emg and should be a game changer. See the Cornell PR and the paper for the details.