Subvocal's glasses read silent mouthing with radio waves, no electrodes on your face

Silent speech input no longer requires electrodes stuck to the face; a pair of glasses uses radio waves to read articulation inside the mouth and turn it into text. The team first taped EMG electrodes to a face at a hackathon and got only letter-level results, then concluded that muscle electrical signals shift with sweat, suffer electrical interference and placement variability, and require wearing electrodes on the face. After switching to RF, the team confirmed a usable signal, then set up booths in their office and paid people to read prompts wearing the glasses, producing a scaling curve in which errors fell quickly as pretraining data grew.

The system gathers information from radio waves that pass through the face and are modulated by the articulators, and a foundation model trained on the collected data decodes those perturbations into text. The route requires an extremely low-noise RF sensing stack and antenna placement whose fields overlap maximally with the tongue, lips and soft palate; the first glasses attempt failed completely, and the team temporarily taped antennas to a face and held them with a head rig. The team designed glasses with antennas embedded in the frames through simulation, and the demo video shows a user who had only a short 20-sentence calibration.

What is shown is a prototype pair of glasses used to collect data in the team's office and in a demo video, where one user had only a short 20-sentence calibration; the team says it will continue R&D over the next few months to turn the prototype into a product.

The team calls cross-user performance genuinely compelling, but that judgment rests on a scaling curve from its own data and a demo video; the material gives no specific error metric or independent evaluation, and silent mouthing takes practice, which the team itself describes as a bit weird to do.