Get started in three steps
- 1Start the cameraAllow access. Good light, hand 30 to 60 cm away.
- 2Teach a signOpen the Teach tab, name it, record 3 or more samples.
- 3Show itThe word appears on the Live tab and is spoken.
Camera is off.
Press Start Camera to begin live tracking.
Press Start Camera to begin live tracking.
Recognised sign
Waiting for a sign
Start the camera, then show a sign you have taught.
Recognition details
Learns only the signs you teach it, so it works with any sign language: Indian, American, British or your own. Not a full translator.
Raw detection
—
Confirmed sign
—
Confidence
0%
Stability timer
0.0s
Teach a sign
Record 3 to 5 samples of each sign. Vary the speed and size a little. Then show it to the camera.
| Label | Type | Samples |
|---|
How it works (demo view)
1 Live Camera Feed
↓
2 Hand Landmark Detection (21 pts × 2 hands)
↓
3 Temporal Tracking + Smoothing
↓
4 Feature Extraction (angles, distances, motion)
↓
5 Universal Sign Recognition (nearest‑neighbor + DTW)
↓
6 Confirmed Text
↓
7 Text‑to‑Speech Output
Advanced / debug
Real-time per-hand geometry + performance internals.
No hands detected yet.
Performance Monitor
Measured FPS0
Frame processing time0 ms
Model inference latency0 ms
Estimated dropped frames0
Camera resolution--
Delegate--
Honesty note: FPS, latency and confidence are measured live on your device; nothing is hard-coded. Recognition only works for signs you record yourself. It is a prototype nearest-neighbour classifier, not a trained production model.