THE ESSENTIALS IN 30 SECONDS

  • Subvocal, in Y Combinator's Fall 2026 batch, says its glasses send low-power radio waves through the face and use models to decode silent mouthing into text.
  • US preorders take a $199 refundable deposit toward a reported $499 early-bird price; first shipments are estimated for Q1 2027, subject to FCC approval.
  • The 200+ words-per-minute figure is the company's claim; it has published no error rate, and its founders say current hardware is still rough.

The Subvocal Company, a three-person San Francisco startup in Y Combinator's Fall 2026 batch, has opened preorders for glasses meant to let people dictate to a computer without making a sound. Co-founder Aarush Yadav announced the preorders and a $1.3 million raise from Y Combinator and Afore Capital on X on October 30, and presented the product on Hacker News on November 2.

How the company says it works

According to the startup's YC launch post, the glasses emit low-power radio waves that pass through the wearer's face and measure how those signals change as the muscles involved in speech move. Models trained on this data then convert the changes into text in real time. There are no cameras. On Hacker News, the founders said the system can also pick up mouthing with closed lips after fine-tuning, and that a demo user needed a short calibration of 20 sentences. They said camera-based lip reading, which they tried first, hit an accuracy ceiling.

The company claims open-ended decoding at more than 200 words per minute and describes a "reasonably low error rate" on a small dataset. It has not published a word error rate, and no independent test has been reported. The founders acknowledge that the current hardware, a semi-portable board paired with glasses, looks worse than the product they plan to ship, and they intend to collect large amounts of RF data to pretrain a model needing only brief per-user calibration.

Price, shipping and open questions

Paid preorders are limited to US addresses. Buyers place a $199 refundable deposit; Digg and a review of the preorder terms by eesel put the early-bird price at $499, while the founders' Hacker News post mentions 500 units with a $200 discount off a final price it does not state. First shipments are estimated for the first quarter of 2027, subject to FCC approval, and the terms say this is not a guaranteed date.

Several basics remain unpublished, including battery life, weight, compatibility with operating systems and apps, and a privacy notice covering the sensor data; eesel notes the current privacy policy covers only the website and preorders. Hacker News commenters questioned whether people will wear dedicated glasses or feel comfortable mouthing words in public. Asked whether RF sensing could be used to eavesdrop, the founder said remote capture is very unlikely because the signal has to pass through the face.

Silent-speech input has mostly been a research topic until now, with prototypes such as Cornell's sonar-based EchoSpeech glasses and MIT's AlterEgo.