Dual-Hand ASL Interpreter — v2.0

HAND
SPEAK

Sign language interpreter that runs entirely in your browser. Both hands tracked, 30+ signs recognised, zero dependencies beyond a webcam.

🤲 Launch Full App ⚡ Try Lite Version ★ GitHub
2
Hands Tracked
30+
Signs Recognised
0
Dependencies
1
HTML File
What it does
BUILT FOR
REAL SIGNING
Most browser ASL tools track one hand. Sign language uses both. HANDSPEAK tracks them simultaneously and combines their positions to recognise two-hand signs you actually use.
🤲
Dual Hand Tracking
Both hands tracked simultaneously via MediaPipe. Each gets its own colour-coded skeleton overlay — yellow right, pink left.
🔤
Full ASL Alphabet
Rule-based classifier covers A–Z static letters using finger curl ratios, spread angles, and pinch geometry.
🗣️
Two-Hand Signs
Dedicated two-hand ruleset for common words: MORE, PLEASE, THANK YOU, HELP, SORRY, FRIEND, WHAT, WHY and more.
⏳
Hold-to-Commit
Hold a sign steady for ~0.7 seconds — a gold arc fills as confirmation. Prevents accidental letter entry from passing gestures.
📝
Sentence Builder
Signs accumulate into a live sentence. Save to history with timestamps, add spaces, backspace, or clear — all without touching a keyboard.
⚡
Zero Setup
Single HTML file. No npm, no build step, no backend. All dependencies load from CDN. Open the file and start signing.
Reference
SIGNS YOU
CAN USE
One-Hand Letters
Two-Hand Words
Under the hood
HOW IT
WORKS
No ML model file to download. Everything runs in pure JavaScript using geometric rules computed from hand landmarks in real time.
01
📷
Webcam Input
MediaPipe Camera Utils captures frames from your webcam at up to 30fps and pipes them into the hand detection model.
02
🦴
21 Landmarks
MediaPipe Hands returns 21 3D landmarks per hand — wrist, knuckles, joints, and fingertips — for both hands simultaneously.
03
📐
Geometric Rules
Per-frame: finger curl ratios, pinch distances, spread angles, and palm direction are computed and matched against hand-crafted sign rules.
04
⏳
Hold & Commit
A sign must be held steady for ~22 frames before it commits to the text output — filtering out accidental or mid-transition frames.
Choose your version
FULL OR
LITE
Lite Version
HANDSPEAK LITE
Camera fills the entire screen, sign overlaid in giant text. Minimal UI — just sign and watch the text appear at the bottom.
Full-screen camera immersive view
Giant sign overlay on camera
Slim output bar at the bottom
Same dual-hand classifier
Half the file size
Launch Lite →