Camera
This tool needs your camera to see your hands.
The video is processed locally and never uploaded.
Reading
How to use it
- Turn on the camera and let your hand fill a good part of the frame — roughly arm's length, well lit, plain background behind it.
- Make a sign with one hand. The skeleton overlay tells you the hand is being tracked.
- Hold it still for about half a second. The bar under the letter fills, and then the sign is added to the transcript.
- Press Speak to hear the transcript read aloud, or edit the text first.
What it can and can't recognise
Letters A–Z covers the fingerspelling alphabet. J and Z are drawn in the air in real ASL — only their handshape is matched here, so they show with a ~ and are unreliable. M, N, T and S are all fists that differ only in where the thumb hides, and depth is the noisiest thing the tracker measures, so expect mistakes between them.
Common signs covers hello, thank you, please, stop, yes, no, help, I love you, okay and you. Several of these are two-handed or carry movement in real ASL; this app matches a one-handed still approximation of each. It is a demo of the recognition pipeline, not a substitute for an interpreter.
The two vocabularies are kept separate on purpose: a fist is both S and yes, and a pointing finger is both Z and you. Switching modes is more honest — and much more accurate — than guessing between them.
Keyboard shortcuts
- Space insert a space in the transcript
- Enter speak the transcript
- Backspace undo the last character
- Esc clear the transcript
Space, Enter and Backspace are ignored while you are typing in the transcript or while a button has focus, so they keep their normal behaviour there. Esc works anywhere.