🖐️ SGT: Smart Gesture Translator

A Real-Time AI System that gives a voice to Sign Language.

Built for Dimension X Hackathon 2026

💡 Overview

SGT (Smart Gesture Translator) is an assistive technology project designed to bridge the communication gap between the mute community and the general public.

Unlike standard sign language detectors that just output letters, SGT uses Generative AI (Google Gemini 1.5 Pro) to intelligently reconstruct broken sentences and Google TTS to speak them aloud in a natural human voice.

🚀 Key Features

• Real-Time Detection: Uses MediaPipe and a custom Deep Neural Network (DNN) to detect hand signs instantly.
• Generative AI Correction: Integrated Google Gemini 1.5 Pro to fix grammar, split words (e.g., "IAMFINE" -> "I am fine"), and correct spelling.
• Natural Voice Output: Uses Google Text-to-Speech (gTTS) for high-quality audio feedback.
• Smart "Safety Net": Zero-latency detection for common phrases like "Hello World" or "Emergency."

🛠️ Tech Stack

• Language: Python
• Computer Vision: OpenCV, MediaPipe
• AI Model: TensorFlow/Keras (Custom trained CNN/DNN)
• LLM Integration: Google Generative AI (Gemini 1.5 Pro)
• Audio: gTTS (Google Text-to-Speech)

📦 Installation & Setup

1. Clone the Repository

   git clone [https://github.com/ironman1947/SGT-Smart-Gesture-Translator.git](https://github.com/YOUR_USERNAME/SGT-Smart-Gesture-Translator.git)
   cd SGT-Smart-Gesture-Translator

2. Install Dependencies

   pip install opencv-python mediapipe numpy tensorflow google-generativeai gTTS playsound python-dotenv

3. Setup API Key (Crucial Step)

   • Create a file named .env in the project root.
   • Add your Google Gemini API key inside it:

     GEMINI_API_KEY=AIzaSy... (Paste your key here)

4. Run the System

   python Real.py

🎮 Controls

• Sign A-Z: Spell your words.
• Press 'C': Confirm a letter.
• Press 'T': Add a space (Optional, AI handles this too!).
• Press 'S': SPEAK (Triggers Gemini AI + Google Voice).
• ESC: Exit.

🤖 AI Model Details

The system relies on a custom-trained Keras model (SGT.h5) trained on a dataset of 26 alphabets + special command gestures.

👥 The Team

• Om Chougule
• Harsh Desai - Team Lead
• Rohan Ilake
• Nikhil Gosai