Open dataset

Help us build the world's most diverse sign-language training set.

Record a few seconds of any sign from your webcam. We extract pose, hand, and face landmarks — never your face image — and add the sample to an open training pipeline that improves real-time recognition for everyone.

50 × 1662 Frames × landmarks per sample
3 Sign systems: PSL · ASL · Custom
100% Private — no faces leave your device
Track Pose, hands, face — 1662 features per frame.
Standardize Resampled to a 50-frame fixed-length sequence.
Train Saved as drive-ready .npy samples and metadata.
Text → Sign Type a phrase, watch every word signed back.

Capture

Hold a sign for 2–4 seconds and press Stop.
REC
0.0s 0 frames
Ready to start
What sign are you recording? 50 × 1662
Reference clip PSL
Type a sign name to preview the canonical clip.
Frames captured 0