SOMEONE MADE DICTATION THAT NEEDS NO MICROPHONE
lipflow: hold a key, silently mouth words at your webcam, they get typed wherever your cursor is
→ runs fully local
→ mac, windows, linux
→ you read 24 sentences and it trains on your face
→ free, MIT
so you can dictate in a silent office or while someone else is talking on a call
i ran the model in a sandbox on cpu. it reads a 3 second clip in about 3 seconds. on random people's clips the guesses were garbled, so the face training step is doing the heavy lifting
solo dev, 431 stars in 4 days