Video Hardcoded Subtitle Extractor

Extract hardcoded/burned-in subtitles from videos using OCR technology. Available as both a desktop application and web interface. This implementation using PaddleOCR for backend OCR task.

🎯 Features

• GUI and web interface options
• Support for MP4, AVI, MOV video formats
• Adjustable frame rate and confidence threshold
• Multiple language support (English, Chinese, Japanese, Korean, Arabic)
• SRT export format. Also supported bilingual subtitles.
• Note: For long video process recommend install GPU version for efficient of speed process (about 1/5 the length of the video)

⚙️ Requirements

• Python 3.8+
• NVIDIA GPU (optional)
• CUDA Toolkit 11.8, 12.0+ (for GPU acceleration)
• 4GB RAM minimum (8GB recommended)

📥 Installation

Option 1: Conda Environment

For GPU version you should install CUDA and cuDNN (version base on their Install paddlepaddle)

# Create conda environment
conda create -n subtitle-env python=3.10
conda activate subtitle-env
# For GPU support (optional)
pip install paddlepaddle # pip install paddlepaddle-gpu==2.6.1 for GPU version
# Install dependencies
pip install -r requirements.txt

Option 2: Docker

# Install NVIDIA Container Toolkit first
# Then build and run with GPU support
docker-compose -f docker-compose.yml build
docker-compose -f docker-compose.yml up

🚀 Usage

Desktop Application

# Launch GUI
python gui.py

Web Interface

# Launch web app
streamlit run app.py

Link Demo