These Glasses Want to Read Your Lips—the Bus Stop Is the Real Test

In the 1960s, NASA researched silent speech in hopes of letting astronauts communicate in noisy environments. More than sixty years later, we are still shouting at our phones. In his tweet, Brian Roemmele put these two things together and casually introduced a new project: Subvocal.

Aarush Yadav has raised $1.3 million from YC and Afore Capital to build a pair of glasses. It uses non-contact sensing to track the internal muscles involved in speech and decode nearly invisible mouth movements into text, with a claimed speed of more than 200 words per minute. The goal is straightforward: write messages, take notes, and use AI without a keyboard or speaking aloud. The website puts it even more bluntly: turn silent speech into text at near-conversational speed.

In this demo, two people sit in chairs having a conversation, and the captions display the line 'Typing was too slow.' The product is trying to claim the space between keyboard and voice input. Keyboards are slow, and talking to a computer is awkward, especially in an office, on the subway, or in a café. Silent mouth movements at least lower the social cost.

The website provides fairly complete information: preorders require a $199 refundable deposit; the early-bird total price is $499, the regular price is $699, and the remaining balance is $300; shipping is expected to begin in Q1 2027; paid preorders are currently open only in the U.S.; prescription lenses are optional at an additional cost; and it uses no camera, nor does it do video or camera-based lip reading. The page also notes that the product is still in development and final specifications are not set.

Under the tweet, congratulations were quickly overshadowed by a question. One user pointed out directly that silent input has been demonstrated for decades, and the hard part is signal quality in everyday environments. Chewing, swallowing, facial hair, and skin oil all interfere with electromyographic (EMG) signals. 200 words per minute in the lab does not equal 200 words per minute on a bus. He wants to see the signal-to-noise ratio data.

This follow-up question is crucial. Silent input's real competitor is the input method. Users won't ask only about peak speed; they'll also ask about error rates, the cost of corrections, how long it can work continuously, and whether wearing a mask or talking will trigger false inputs. If 200 words per minute holds only in a quiet room, it looks more like a demo than a tool. The website's inclusion of 'no camera' in the FAQ also shows that privacy concerns are an unavoidable threshold for this kind of glasses. It reads muscle signals, which is still far from mind reading, but users will still worry about what the device can record and upload.

Some people in the replies are already getting ready to place an order, while others have posted the 'shut up and take my money' meme. The excitement is understandable. The glasses form factor is lighter than a headset, more discreet than a microphone, and closer than a keyboard to the fantasy of 'think it and write it.' Brian's line that 'the interface has always been on our face' is elegant, but the face is also busy. Eating, yawning, and making expressions all add noise to muscle signals.

Subvocal's preorder link is now live: http://subvocal.co/preorder?utm_source=x . The $199 deposit is refundable, lowering the cost of trying it out, but the shipping timeline may change and final specs are not set. What you're buying now is a roadmap; the finished product is not finalized.

NASA wanted to solve communication in noise sixty years ago; today these glasses want to turn mouth shapes into text. The direction is interesting. First, wait for a real-world bus-stop test: if it can reliably read 200 words per minute amid shaking, noise, and other people's gazes, the keyboard will lose a friend; if not, it at least reminds us that the face has always been an interface—it just wasn't charging before.