You articulate the words without voicing them. The muscles that would have produced speech still fire, and Somach reads them at the skin and turns them into text.
You form the word and stop short of voicing it. Lips closed, no air, no sound. The motor commands still reach the muscles of articulation.
Surface electrodes read that muscle activity at the skin, the same class of sensor as a consumer EMG fitness band. Nothing implanted. Nothing invasive.
A model running on your own device decodes the pattern into words. The signal never has to leave you for the sentence to arrive.
A benchtop prototype, a corpus of 4,033 recorded silent-speech samples from a single speaker, and a closed-vocabulary decoder that works. Also a long list of things that do not work yet.
Somach Systems is one person in San Francisco. This is a research bet, not a product with a ship date. The waitlist is for people who want to watch: researchers, people who cannot rely on their voice, and anyone who has wanted to answer a message without saying it out loud.
On the bench: widening the analog front end to 20–450 Hz. The old corpus was recorded through a 40 Hz roll-off, so new hardware means a new corpus.