Subvocal Smart Glasses: Silent Speech Recognition via RF Sensing

Core Technology: RF-Based Silent Speech Recognition

The Subvocal Company is developing smart glasses that allow users to dictate text or command computers without speaking aloud. The system uses embedded radio frequency (RF) antennas to detect subtle movements of the tongue, lips, and soft palate. Unlike traditional EMG-based systems that require skin-contact electrodes, this RF-based approach is non-invasive and functions by measuring how articulatory movements modulate radio waves as they pass through the face. The company trains foundation models on this RF data to map specific signal perturbations to text, achieving performance that allows for continuous dictation.

Engineering Challenges and Development

Initial development focused on overcoming the limitations of electromyography (EMG), which proved unreliable due to signal interference from sweat, electrode placement variability, and user discomfort. The team transitioned to RF sensing, which required significant R&D to solve signal noise and antenna placement issues. Because facial anatomy varies significantly between individuals, the team addressed cross-user generalization by scaling their pretraining dataset, collecting thousands of hours of data from diverse participants. The current prototype, which features antennas embedded within the glasses frames, is designed to be a consumer-ready product by Q1 2027.

Community Feedback and Industry Context

Public reception on Hacker News has been mixed, balancing excitement for the potential of hands-free, private computing against concerns regarding hardware form factors and privacy.

Key User Concerns

Form Factor and Fit: Many users noted that glasses are notoriously difficult to fit, with commenters requesting alternatives like clip-on modules for existing eyewear or non-glasses wearables to avoid the "one-size-fits-all" issue.
Privacy and Security: Commenters raised questions about whether the system operates locally or via the cloud, with strong preferences for offline-only processing to prevent potential misuse of sensitive speech data.
Health and Safety: There is ongoing discussion regarding the long-term health implications of having RF-emitting hardware in close proximity to the eyes and brain, with some users calling for formal medical safety validation.
Accuracy and Utility: While the technology is viewed as a breakthrough for accessibility and public-space productivity, potential users expressed skepticism about the current accuracy levels, noting that the website lacks specific performance metrics.

Expert Insights and Comparisons

Community members compared the project to existing research, such as the MIT Media Lab's AlterEgo and various open-source lip-reading projects. Some users highlighted the potential for "internal speech"—where the user does not move their lips at all—as a future milestone for the technology, though they noted this would likely require significant user training and real-time feedback loops.

"I expect training internal speech might involve more real-time feedback to help the user find the right degree and kind of articulation. I realize you're still in the prototype phase but I assume an increasing number of people are getting multiple longer sessions." — @mrandish

"I'm more looking for the Jane from Speaker for the Dead version. In-ear device that can hear and feel the subvocalization." — @thisisauserid

Ultimately, the project is viewed as a significant step toward solving the "public dictation" problem, though its success will likely depend on achieving high accuracy, ensuring user privacy, and providing a comfortable, inclusive hardware design.