An era is arriving where text can be typed simply by moving the mouth without making a sound. US startup Subvocal has unveiled muscle-recognition typing glasses with an input speed exceeding 200 words per minute.
Subvocal announced the launch of pre-orders alongside news of securing $1.3 million in funding from Y Combinator and Afore Capital on October 30 local time. The early-bird price is set at $499 (approx. 670,000 won), with the product page listing a regular price of $699 (approx. 940,000 won).
Subvocal is a system that reads speech muscle movements using a radar-based radio-frequency approach without cameras. Radio waves spread through the air, detecting movements beneath the skin even without direct contact. Subvocal stated that the system only recognizes words deliberately mouthed and does not read thoughts. In demonstration videos, rapid typing was achieved simply by slightly mouthing words.
The project originated at Stanford University's hackathon, TreeHacks. Subvocal shared in an X post that they attached electromyography (EMG) sensors to the face. The model was trained within 24 hours using data from silently pronouncing the 26 letters of the alphabet, achieving a character classification accuracy of around 80%.
They subsequently tested four methods: EMG, video, Inertial Measurement Units (IMUs), and radio waves. EMG requires skin contact, and its accuracy fluctuates significantly with each use. Webcam video brought the word error rate down to 17%, but since it only captures the lips, distinguishing sounds with similar mouth shapes like 'P' and 'B' proved difficult. IMUs only capture full-facial movements, presenting clear limitations as vocabulary scales up. Conversely, Subvocal claims that its current prototype glasses allow first-time users to calibrate in about one minute to read continuous speech, achieving 85–92% word accuracy.
As the foundation for their radio-wave method, they cited a 2022 paper on radar-based silent speech recognition published in the academic journal Scientific Reports.
Devices targeting silent speech input have appeared before. On January 29, Apple acquired Israeli startup Q.ai, which reads silent speech through minute movements of facial skin, for $2 billion. In September last year, AlterEgo unveiled a face-attached wearable originating from research at the Massachusetts Institute of Technology (MIT) Media Lab and demonstrated videos of typing without speaking.
The industry appears to be focusing attention on silent speech recognition, which eliminates the need to type by hand or make sounds out loud.
Initial shipments of Subvocal's glasses device are expected in Q1 2027. Subvocal added that the timeline may change as the product remains in development.
