Subvocal Debuts YC Glasses That Decode Silent Mouthing to Text

Subvocal's glasses emit radio waves that pass through the wearer's face. As the signal crosses the tongue, lips and soft palate while someone silently mouths words, it becomes modulated, and foundation models decode those perturbations into text.

"They emit radio waves, similar to Bluetooth, that pass through your face," the founders write in their Show HN post. "We train foundation models on this data to decode these perturbations to text."

The company, The Subvocal Company, launched this week out of Y Combinator's F26 batch. Its founders, Aarush and Arnav, studied computer science at Georgia Tech and began the project at Stanford's TreeHacks hackathon in February, originally using EMG sensors on the face before pivoting to RF.

Published performance numbers are thin. The promotional video shows roughly a 5% word error rate, end-to-end latency "under 500 milliseconds typically," and a 20-sentence calibration for new users. The founders also say the system works for "internal speech" with closed lips, "though it requires practice." There is no independent benchmark.

Preorders for the first 500 units are open at $200 off the eventual retail price, with shipping targeted for Q1 2027 and English the only supported language at launch. Allocation, the company says, is "running out quickly." The post climbed to 323 points and 211 comments within hours of going up, and it sits sideways to the 38 voice-AI stories we've logged in the last 90 days, most of them transcription, TTS, or voice agents — the input here is motion, not sound.