Subvocal
A silent speech input device in the form of smart glasses that uses radio-frequency sensing of articulatory mouth movements to convert silently mouthed words into text.
Timeline (last 90 days)
Subvocal is a company founded by two Georgia Tech dropouts; its product is smart glasses that use radio-frequency sensing to detect a user's silent mouth movements and decode them into text.
The Subvocal team initially replicated MIT Media Lab's AlterEgo approach at the Stanford TreeHacks hackathon, using EMG electrodes to read electrical signals from facial muscles; the 36-hour prototype barely worked at the letter level and did not win an award.
The EMG approach has hard limitations such as sweating changing the signal, severe electrical interference, sensitivity to electrode placement, and users' reluctance to stick electrodes on every day, which prompted the team to switch to a radio-frequency approach.
Subvocal's radio-frequency approach embeds antennas in the glasses frame that emit Bluetooth-like radio waves through the face; when the waves pass through the tongue, lips and soft palate they are modulated by silent mouth movements and then decoded into text by a foundation model.
The Subvocal system does not require the user to move their lips or open their mouth; it can be fine-tuned for "inner speech" in which the user shapes words with the mouth closed, which users can master with practice.
In the demo video, YC's Dave was able to complete input with the Subvocal glasses after calibrating with only about 20 sentences.
Cross-user generalization is the toughest problem Subvocal faces; by paying people in the office to wear the glasses and read prompts aloud to collect data and by scaling up pretraining, the team found that the error rate drops rapidly as data volume increases.
Subvocal is currently at the prototype stage, and the team plans to ship in the first quarter of 2027.
The first 500 units of Subvocal are already available for pre-order, with a $200 discount off the final price.
The advantages of radio-frequency silent speech are that it is completely contactless and requires no skin electrodes, making it easier to build into everyday wearables; the challenges include the sensitivity of radio-frequency signals to head posture and the environment, consistency of long-term wear, and adaptation to different languages and accents.
1 more timeline events
All Facts (11)
Subvocal is a company founded by two Georgia Tech dropouts; its product is smart glasses that use radio-frequency sensing to detect a user's silent mouth movements and decode them into text.
50%UnverifiedThe first 500 units of Subvocal are already available for pre-order, with a $200 discount off the final price.
50%UnverifiedThe advantages of radio-frequency silent speech are that it is completely contactless and requires no skin electrodes, making it easier to build into everyday wearables; the challenges include the sensitivity of radio-frequency signals to head posture and the environment, consistency of long-term wear, and adaptation to different languages and accents.
50%UnverifiedThe team reflected on first principles: solving the whole problem top-down failed completely, and it only worked once the antenna, electronics and machine learning components were broken down to the most basic level and optimized one by one.
50%UnverifiedThe EMG approach has hard limitations such as sweating changing the signal, severe electrical interference, sensitivity to electrode placement, and users' reluctance to stick electrodes on every day, which prompted the team to switch to a radio-frequency approach.
50%UnverifiedSubvocal's radio-frequency approach embeds antennas in the glasses frame that emit Bluetooth-like radio waves through the face; when the waves pass through the tongue, lips and soft palate they are modulated by silent mouth movements and then decoded into text by a foundation model.
50%UnverifiedThe Subvocal system does not require the user to move their lips or open their mouth; it can be fine-tuned for "inner speech" in which the user shapes words with the mouth closed, which users can master with practice.
50%UnverifiedCross-user generalization is the toughest problem Subvocal faces; by paying people in the office to wear the glasses and read prompts aloud to collect data and by scaling up pretraining, the team found that the error rate drops rapidly as data volume increases.
50%UnverifiedThe Subvocal team initially replicated MIT Media Lab's AlterEgo approach at the Stanford TreeHacks hackathon, using EMG electrodes to read electrical signals from facial muscles; the 36-hour prototype barely worked at the letter level and did not win an award.
50%UnverifiedSubvocal is currently at the prototype stage, and the team plans to ship in the first quarter of 2027.
50%UnverifiedIn the demo video, YC's Dave was able to complete input with the Subvocal glasses after calibrating with only about 20 sentences.
50%