A San Francisco startup is developing AI-powered glasses that can turn silently mouthed words into text, offering a potential alternative to typing or speaking aloud at work.
The Subvocal Company unveiled the device in late October 2026, claiming its prototypes can process more than 200 words per minute, although independent testing has yet to establish how accurately they perform.
The idea is that instead of dictating an email or tapping out a lengthy prompt, users mouth the words they want to communicate while wearing the glasses. The technology is designed to detect the physical signals associated with those movements and use artificial intelligence to convert them into text.
However, it does not claim to read thoughts.
How the AI Glasses Interpret Silent Speech
The Subvocal Company says its glasses use low-power radio-frequency sensing to detect subtle changes as someone deliberately forms words without speaking.
Co-founder Aarush Yadav told the New York Post that the system emits radio waves at a frequency similar to Bluetooth, but at a much lower power.
As those waves interact with facial muscles and other structures involved in articulation, the signals change. The company's AI models then analyse those changes and attempt to reconstruct the words.
Subvocal advertises speeds of more than 200 words per minute, potentially allowing users to compose emails, dictate notes and give instructions to AI tools without making a sound. That figure, however, remains a company claim.
The startup has not published independent accuracy results, so it is not yet clear how reliably the glasses would handle everyday conversations, long messages or unfamiliar words.
The founders have also demonstrated a proposed feature that would allow people wearing the glasses to communicate silently with one another across a room. The demonstration has attracted attention online, but it does not establish how consistently the feature works outside a controlled setting.
Yadav and fellow co-founder Arnav Kumar began developing the technology at Stanford University's TreeHacks hackathon in February 2026. Both are computer science students at the Georgia Institute of Technology and have taken leaves of absence to work on the startup.
NEW: Subvocal built smart glasses that let you talk to your computer without making a sound.
— Ritwik Pavan (@ritwikpavan) October 30, 2026
The glasses turn deliberately mouthed words into text, so you can write messages or talk to AI without speaking out loud or touching a keyboard.
-Company says prototypes can exceed… pic.twitter.com/WJzEGMJ8cH
Privacy Remains a Question
Yadav sees the technology as a way to make computer interaction faster and less intrusive. Voice dictation can be convenient, but speaking to a computer is not always practical in a shared office, library or public space.
The founders are also targeting people who regularly use AI tools. Workers who send lengthy prompts or manage several AI agents could use silent speech to issue instructions without repeatedly typing or talking aloud.
While the company says the glasses do not require a camera for silent-speech input, it does not answer every question about privacy. The device relies on physiological signals, and the handling of that data will matter as much as the absence of a camera.
Columbia University neuroscientist Rafael Yuste told the New York Post that he was intrigued by the engineering but cautious about its implications. He noted that Subvocal had not published accuracy figures, making it too early to describe the technology as a breakthrough.
Yuste also warned that collecting more physiological data could create privacy concerns, particularly information were shared with companies or governments. His comments identify a potential risk, not evidence that Subvocal is misusing user data.
Questions remain over where the radio-frequency signals are processed, whether raw sensor data are retained and whether users can opt out of contributing information to model training.
He raised another concern about the effect of replacing familiar ways of communicating. Writing and articulating words can help people organise and refine their thoughts, he said, and relying on silent digital input could affect those processes in ways that are not yet understood.
Subvocal's AI Glasses Are Available to Preorder
The glasses are not yet a finished consumer product.
Subvocal is accepting US preorders at an early-bird price of $499 (£372), reduced from its listed $699 (£521) price. Customers initially pay a refundable $199 (£148) deposit, with the remaining $300 (£224) due later. Prescription lenses cost extra.
The company expects shipping to begin in the first quarter of 2027, although that is an estimate rather than a guaranteed delivery date. Its specifications and design are still being developed, and the schedule could change.
Yadav has acknowledged that the team is continuing to improve the glasses' accuracy and design, describing the device shown in its demonstration video as an early prototype. Whether it can become a reliable workplace tool will depend on how well the technology performs beyond its demonstrations.
