SamuelParker/Malaysian-Sign-Language-Recognition

Malaysian-Sign-Language-Recognition

Real-time Malaysian Sign Language Alphabet Detection using MediaPipe, PyTorch & Flask

A web-based AI system that recognizes Malaysian Sign Language (MSL) hand gestures from your webcam and converts them into live text with stable prediction logic, plus optional text-to-speech output. This project was built for a hackathon to demonstrate accessible communication using computer vision and deep learning.

Features 🔹 Real-Time Hand Tracking

• —Uses MediaPipe Hands to extract 3D landmarks from video frames with high accuracy.

🔹 Deep Learning Classification

• —A custom PyTorch fully connected neural network predicts the MSL alphabet from extracted features.

🔹 Stability Logic (Noise Reduction)

• —A 70% rolling prediction buffer is applied to ensure only stable letters are added to the final output.

🔹 Web-Based Interface (Flask + SocketIO)

• —Live webcam feed
• —Real-time predictions
• —Stable text output
• —Interactive controls (Clear, Delete, Space, Read Aloud)

🔹 Speech Output (Text-to-Speech)

• —Converts the recognized word/sentence to speech using pyttsx3.

LIVE AT :

Installation for you to run on local :

1. 1.Clone the repo : git clone https://github.com/yourusername/MSL-SignLanguage-Recognition.git cd MSL-SignLanguage-Recognition
2. 2.Install dependencies : pip install -r requirements.txt
3. 3.Run the app : python app.py
4. 4.Open in browser : http://localhost:5000

Future Improvements 1.Full word-level recognition 2.Real-time sentence translation 3.Dataset expansion for 2-hand gestures 4.Mobile version (TensorFlow Lite) 5.Multi-language speech output

Credits Developed by Samuel Raj. For educational and accessibility innovation.