ASL Fingerspelling Recognition
Client-Side OnlyLive ASL alphabet letter recognition from your webcam — fingerspelling, not translation
Hold up one hand, fingerspelling a letter of the ASL alphabet, facing the camera. Recognition runs entirely in your browser (MediaPipe HandLandmarker + a k-NN classifier trained on real photos) — no video ever leaves your device.
Fingerspelling only — not sign-language translation. Supports 24 static letters (A-Z excluding J and Z, which require motion a single frame can't capture). Measured held-out accuracy: 79% — real, useful, but not perfect; see the User Guide for known letter mix-ups.
Supported letters
ABCDEFGHIKLMNOPQRSTUVWXY
J and Z are excluded — both require a traced motion in real ASL, indistinguishable from other letters (or each other) in a single static frame.
What this doesn't do: not a sign-language translator — it recognizes individual finger-spelled letters, never whole signed words or ASL grammar. The 79% accuracy is measured on real held-out photos with plain backgrounds, not verified end-to-end on a real live webcam feed in this environment — visually similar hand shapes (e.g. U/V/R, M/S/N/A) are a real, disclosed source of mix-ups, not a bug.