Every hand
has a voice.
SignLang AI translates Indian Sign Language into text and bilingual speech, in real-time through your webcam. No special hardware. No downloads. Just show your hand.
Built for real accessibility
Every feature is designed to remove communication barriers for India's deaf and hard-of-hearing community.
Two-hand detection
MediaPipe tracks both hands simultaneously — both skeletons render on the video overlay. The dominant hand drives the AI prediction, capturing the full range of ISL gestures.
0199.9% LSTM model
A 3-layer PyTorch LSTM trained on 42,000+ landmark sequences across 35 ISL classes. Majority-vote buffer smooths predictions for stable, flicker-free output.
02Bilingual TTS
Every recognized word and full sentences are spoken aloud via gTTS — in both English and Hindi. Toggle languages on the fly from the dashboard.
03NLP sentence polishing
Raw signs are automatically stitched into grammatically clean sentences. Number words convert to digits, consecutive duplicates are removed, and sentences are properly capitalised and punctuated — all in real-time.
04Hindi translation
One click translates the full recognized sentence into Hindi, displayed inline on the dashboard and speakable via voice output.
05Confidence sparkline
A live chart plots prediction confidence over time so you can see detection quality at a glance and adjust your hand position for optimal accuracy.
06Sign dictionary
Browse all 35 supported ISL signs in a slide-up reference drawer, organized alphabetically — letters A–Z and numbers 1–9. Perfect for learning while using.
07Copy & share
Copy the full recognized sentence to clipboard in one click, share session results, and download professional PDF reports with word logs and timestamps.
08Keyboard shortcuts
Space to start/stop, C to clear, H to toggle Hindi, D for dictionary, ? for the shortcuts overlay — full keyboard accessibility built in.
09How it works
Webcam capture
OpenCV streams 30 FPS from your camera. Works on any laptop or phone with a front camera.
Landmark extraction
MediaPipe detects 21 3D key-points per hand across both hands simultaneously at real-time speed.
LSTM classification
A 30-frame landmark sequence is classified by a 3-layer LSTM achieving 99.9% validation accuracy.
NLP + Voice output
Signs are polished into sentences with NLP, displayed on screen, and spoken in English or Hindi.
Start translating sign language today.
Free account. No downloads. No hardware. Just open your camera and start signing.
Create Free Account →