OMG (Online Meeting Guru)

English | 简体中文

OMG is a tool for processing meeting recordings (or screen captures), extracting key frames, and generating transcripts. It features a lightweight modern interface built with Gradio.

Features

• Two input modes:
  • Video file upload: Process pre-recorded videos
  • Screen capture: Capture and process screen content in real-time
• Extract key frames based on content similarity
• Generate transcripts using OpenAI's Whisper models
• Support manual editing of extracted frames and transcript before export
• Export results as PDF, audio, and text files
• Lightweight modern interface built with Gradio
• Extreme compression, successfully reducing a 2.48GB video to just 321MB

Requirements

• Python>=3.8 (tested on 3.12)
• CUDA-compatible GPU (optional, for faster processing)
• FFmpeg (for audio processing)

Installation

1. Clone the repository:

git clone https://github.com/ZhenghaoYang19/omg.git
cd omg

2. Create virtual environment and install dependencies:

   Using uv (Recommended)

   # Install uv (or download from https://github.com/astral-sh/uv)
   pip install uv
   
   # Create and activate virtual environment
   uv venv
   .venv\Scripts\activate  # On Windows
   source .venv/bin/activate  # On Linux/macOS
   
   # Install dependencies
   uv pip install -r requirements.txt

   Using pip

   pip install -r requirements.txt

3. Install FFmpeg:

• Windows: Download from FFmpeg website
• Linux: sudo apt-get install ffmpeg
• macOS: brew install ffmpeg

Usage

1. Start the web interface:

python app.py

2. Open your browser and navigate to http://localhost:7860

3. Choose your input mode:

   Video Upload

   • Upload your video file
   • Adjust configuration settings (optional)
   • Click "Process Video"
   • Wait for processing to complete
   • Switch to "Results & Export" tab to view results

   Screen Capture

   • Select capture type (Monitor or Window)
   • For Monitor capture: Choose the monitor from the dropdown
   • For Window capture: Enter the window title (or part of it)(e.g. chrome, edge, TencentMeeting)
   • Adjust configuration settings (optional)
   • Click "Start Capture"
   • Click "Stop Capture" when finished
   • Switch to "Results & Export" tab to view results

4. Export results:

   • Select desired export options (PDF/Audio/Transcript)
   • Click "Export Selected Files"
   • Download the exported files

Configuration

You can adjust the following parameters:

• Similarity Threshold (0.0-1.0): Controls how different frames need to be to be considered key frames, suggested value between 0.6 and 0.7
• Frames Per Second: Number of frames to sample per second, suggested value between 0.2 and 5
• Start/End Time (Video upload only): Process only a specific portion of the video
• ASR Model: Choose between different Whisper models
• ASR Device: Select processing device (suggested: auto)
• Frame Comparison Method: Choose between different frame comparison algorithms

Default values can be modified in config.json.

Project Structure

omg/
├── app.py              # Web interface
├── omg.py        # Core processing logic
├── utils/
│   ├── compare.py      # Frame comparison functions
│   └── images2pdf.py   # PDF generation
├── config.json         # Configuration file
├── requirements.txt    # Python dependencies
└── output/            # Processing results

Output Directory Structure

For video upload:

output/
└── video_name/
    ├── images/        # Extracted frames
    ├── audio.wav      # Extracted audio
    ├── transcript.txt # Generated transcript
    └── slides.pdf     # Generated PDF (optional)

For screen capture:

output/
└── screen_capture/
    └── YYYYMMDD_HHMMSS/  # Timestamp of capture
        ├── images/        # Captured frames
        ├── audio.wav      # Recorded audio
        ├── transcript.txt # Generated transcript
        └── slides.pdf     # Generated PDF (optional)

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Acknowledgments

• wudududu/extract-video-ppt for the inspiration and reference
• Gradio for the web interface
• OpenAI for the Whisper ASR model