Aarush and Arnav, founders of The Subvocal Company from Y Combinator F26, are building smart glasses that enable silent speech interaction. Their technology allows users to mouth words discreetly with lips closed, which the glasses decode into text at the speed of speech using RF sensing. The concept originated from a Stanford hackathon project where they initially experimented with EMG electrodes before switching to radio waves.
Despite initial challenges with signal interference and user comfort, they developed a working prototype using foundation models to decode RF perturbations. The team overcame cross-user generalization issues by massively scaling pretraining data through paid office booth sessions. The glasses offer a discreet alternative to voice commands, particularly useful in libraries and public spaces where traditional voice tools are impractical.
Their solution represents a significant engineering achievement in silent speech technology, moving from crude prototypes to a functional product that could transform human-computer interaction.
Source: Hacker News · Summarized by HeadlinesBriefing