Subvocal Develops Smart Glasses for Silent Speech-to-Text Input Using RF Sensing

Silent Speech Technology Development

The Subvocal Company, founded by Aarush and Arnav, is developing smart glasses that enable silent speech-to-text conversion. The system allows users to mouth words without vocalizing them, and the glasses decode these movements into text at speech speed. This technology is intended for discreet computer interaction in environments like libraries or offices.

From Hackathon Idea to Product

The concept originated at Stanford’s TreeHacks, where the founders, then CS students at Georgia Tech, built a prototype using EMG electrodes. This initial attempt, inspired by MIT’s AlterEgo, demonstrated the feasibility of silent speech at a letter level, convincing them to pursue the idea further despite its rudimentary nature.

Transition to RF Sensing

The team initially used EMG (electromyography) but encountered issues with signal variability due to sweat, electrical interference, placement inconsistency, and user discomfort. They subsequently switched to RF sensing. The current glasses emit radio waves that pass through the face, and these waves are modulated by the tongue, lips, and soft palate during silent mouthing. Foundation models are then trained to decode these modulations into text.

Engineering Challenges

Developing the RF-based system presented significant engineering challenges. This included building a low-noise RF sensing stack and precisely placing antennas to maximize overlap with articulators. Early attempts to integrate the technology into glasses failed, requiring further development and iteration.