Subvocal
Subvocal is building glasses that decode silent, mouthed speech into text so people can type by speaking without making
The Problem
People often need to turn thoughts into text in places where speaking aloud is awkward or impossible: on a crowded train, in a quiet office, around strangers, or next to coworkers. Today the only option is voice dictation, which requires speaking audibly and exposes private messages, work notes, or instructions to an AI agent to anyone nearby. This makes voice input unusable in many everyday settings, forcing people back to slow manual typing or simply not sending the message at all.
The Solution
Subvocal is a pair of smart glasses that decodes subtle, silent mouthing of words into text in real time, without any sound being produced. The company claims its current prototypes already decode open-ended silent speech at 200+ words per minute and that accuracy improves as more data is collected and models get larger. It is designed as an input method rather than a standalone app, intended to work inside existing tools like ChatGPT, Claude, Cursor, Slack, email, and messaging. A short personal calibration step adapts the system to how an individual articulates words, and the product is explicitly built to trigger only on deliberate articulation rather than continuous listening or recording.