System Information
How Sign-Bridge AI Works
Powered by MobileNetV2 and MediaPipe Hands for real-time ASL recognition
Model Architecture
- Base: MobileNetV2 (Fine-Tuned layers 100–155)
- Input Shape: 128 × 128 × 3
- Output Classes: A–Z + SPACE + DOT + AUTOCORRECT
- Feature Vector: n=1280 (flattened)
- Classifier: Fully Connected → Softmax
Recognition Settings
- Confidence Threshold: 0.75
- Smoothing Window: 10 frames
- Letter Hold Duration: 1.0 sec
- Special Gesture Hold: 1.2 sec
- WebSocket Frame Rate: adaptive
Live System Status
WebSocket Connecting …
ML Model —
Autocorrect —
TTS Engine gTTS Cloud ✓
Tech Stack
Python 3.10
TensorFlow / Keras
MobileNetV2
MediaPipe Hands
OpenCV Headless
FastAPI + Uvicorn
WebSocket
gTTS Cloud TTS
LanguageTool NLP
Browser Camera API
ASL Alphabet Chart
MobileNetV2 Architecture
Special Gesture Commands
- Flat Hand (1.2s hold) → Insert SPACE
- Pinch gesture → End sentence + Speak
- Thumbs Up → Trigger Autocorrect
- Q–T keys → Select word suggestion
- 1–5 keys → Pick autocorrect variant
- Z / Backspace → Delete last letter
- C key → Clear all text