YC funded a startup $1.3M to build silent speech – a dev on x built it in 9mins

Subvocal, founded by Aarush Yadav and backed by Y Combinator and Afore Capital, is developing glasses that convert silent mouthing into text. The company says its glasses track the facial muscles involved in speech and can capture more than 200 words per minute, aiming to make typing or speaking to computers unnecessary. The product has opened pre-orders, but its speed and performance have not been independently demonstrated. The idea also drew skepticism about whether mouthing words in public would feel more awkward than talking aloud.

Developer Joe Blau responded by releasing Subvox, a free, open-source experiment for AirPods Pro 3 that he says he built in nine minutes with a single prompt. It uses earbuds’ head-motion sensors to recognize jaw and head movements associated with a small set of words users train themselves, rather than detecting speech-muscle activity. Setup instructions call for practicing five words ten times each; the tool has no language model and publishes no accuracy results, so it is not a general-purpose dictation system. The quick demo fueled claims that Subvocal has no defensible advantage, but the two efforts solve different problems: Subvox tests a limited vocabulary using existing hardware, while Subvocal promises continuous, much faster silent speech through custom glasses. The key question is whether the glasses can deliver their stated performance in real-world use.