Live ASL recognition.
Built on the CLERC dataset.
One annotated ASL dataset, built with Deaf signers. The demos below are examples of what the data makes possible. CLERC ships the data layer, not the tools.
Open sample on Hugging Face · CLERC-DATA/epeeLive ASL recognition, right in your browser
This page is a live American Sign Language (ASL) recognition demo. Point your camera, sign a word, and the model reads it in real time — no app, no account, no upload. Everything runs on-device in your browser, powered by the Gallaudet model trained on the CLERC sign language dataset, built with native Deaf signers.
It is also a simple way to learn your first ASL signs online: in the education demo you watch a real Deaf signer perform a sign, copy it in front of your camera, and get instant feedback. The demo exists to show what clean, Deaf-built sign language training data makes possible — CLERC ships the data layer. Read the benchmark write-up, see how it compares in the sign language dataset landscape, or explore the open sample on Hugging Face.
What is live ASL recognition?
Sign language recognition (SLR) is AI that reads signs from video. Here, a computer-vision model tracks your hands, body and face through your camera and matches your signing against ASL signs it learned from the CLERC dataset — live, as you sign.
Do I need to know sign language to try the demo?
No. The education demo shows you a reference video of a native Deaf signer, you copy it in front of your camera, and the model scores your attempt. It is designed as a first contact with ASL — no experience needed.
Is my camera video recorded or uploaded?
Your video is never recorded and never uploaded. The camera is read in your browser and turned into anonymous skeleton points — no image, no face. Those points are sent to our server to compute the word, then discarded: nothing is stored. If you choose to tell us a guess was wrong, we keep only the two words involved, and only after you opt in.
Which signs can the demo recognize?
The current vocabulary is 50 ASL signs, recognized one word at a time. It is an early, honest benchmark: about 82% top-1 on held-out corpus video at this vocabulary size, and up to 71% on a signer the model had never seen, measured on the 15-sign version of this demo. In a live browser expect less: lighting, framing and distance from the corpus all cost accuracy. The vocabulary grows with the dataset.
Can I learn ASL online with this demo?
You can learn your first signs. The demo is not a full ASL course — sign language is a rich 3D language best learned from Deaf teachers — but it gives instant, private practice feedback based on reference videos from real Deaf signers.
What dataset powers the model?
The CLERC dataset (Épée): ASL video recorded and annotated by native Deaf signers, with gloss labels, intents and QA validation. A public sample is available on Hugging Face (CLERC-DATA/epee).
Which hand do you sign with?
We tune the camera to you before calibration.