Subvocal raised $1.3M. A developer recreated it with AirPods in 9 minutes
Subvocal, a macOS prototype, has raised $1.3 million to explore the potential of turning silently mouthed words into text using AirPods Pro head-motion data. The prototype does not utilize a language model but instead employs a small-vocabulary template recognizer and a statistical test to determine the authenticity of the signal.
To use the prototype, users must grant Motion permissions on first launch and ensure their AirPods Pro or another headphone that supports head tracking is connected as the Mac's audio output. The mic and speech prompts will only appear once the mic is activated. Users can record sessions using the Record Session button in the toolbar or on the Sensors screen, with recordings saved to the user's Documents folder.
The prototype suggests running the same 5-word script under Normal speech, Whisper, and Silent mouthing conditions, with 10 repetitions each. After recording, users can compare Evaluate results across feature sets in the Train mode, with the dataset stored at ~/Library/Application Support/Subvox/dataset.json.
Brief written by urgent.news from Hacker News's own syndicated text. Machine-written — may contain errors; check the original before relying on it.
