Subvocal: A New Frontier in Silent Speech Technology

Aarush and Arnav, founders of The Subvocal Company, have introduced a novel solution for users who wish to interact with their computers without speaking aloud. Their project, ‘Subvocal,’ utilizes smart glasses equipped with radio frequency (RF) technology to decode silent mouthing into text at the speed of natural speech. This innovation aims to provide a discreet way for users to utilize voice dictation tools in public spaces like libraries or offices where traditional speech is impractical.

The journey to this prototype began at Stanford’s TreeHacks, where the founders initially experimented with electromyography (EMG) sensors. However, they quickly encountered significant hurdles, including signal interference from sweat and the social awkwardness of attaching electrodes to one’s face. After pivoting to RF technology, the team developed a system that emits radio waves through the facial structure. These waves are modulated by the movement of the tongue, lips, and soft palate, allowing the glasses to capture silent speech patterns.

Developing the technology required overcoming immense engineering challenges, particularly in antenna placement and signal processing. The team utilized deep simulations and extensive pretraining data to ensure the system could generalize across different facial anatomies. By scaling their dataset through controlled testing, they achieved a level of accuracy that makes the technology genuinely compelling. The founders emphasize that breaking down complex components—antenna design, electronics, and machine learning—to their fundamental principles was key to their success. With preorders now open for their first batch of 500 units, the team is focused on further research and development, aiming to deliver a polished product by early 2027.

Sources: Hacker News

Aggregated summary. Original: Hacker News