AI Meeting Transcription and Summarization

A tool for transcribing and summarizing meetings from video files or YouTube URLs. This application uses state-of-the-art AI models to perform speaker diarization (identifying who is speaking) and speech-to-text transcription, followed by automatic summarization of key points.

Features

• Process video files or YouTube URLs
• Extract audio automatically
• Identify different speakers (diarization)
• Transcribe speech to text with Whisper models
• Generate meeting summaries with key points
• Support for Spanish and English languages
• Export transcriptions in SubViewer format (.sub)

Requirements

• Python 3.10+
• CUDA-compatible GPU (recommended for faster processing)
• Hugging Face account with API token

Installation

1. Clone this repository:

   git clone https://github.com/matiaszanolli/AI-Meeting-Summary-SPANISH.git
   cd AI-Meeting-Summary-SPANISH

2. Create a virtual environment and install dependencies:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   pip install -r requirements.txt

3. Create a .env file in the project root with your Hugging Face token:

   HUGGINGFACE_AUTH_TOKEN=your_token_here

Usage

1. Start the web interface:

   python web-ui.py

2. Open your browser at http://localhost:7860

3. Upload a video file or enter a YouTube URL

4. Configure the parameters:

   • Select language (Spanish, English, or auto-detect)
   • Choose Whisper model size (larger models are more accurate but slower)
   • Adjust collar value for speaker diarization
   • Enable/disable summary generation

5. Click "Iniciar" to start processing

6. View the transcription and summary in the respective tabs

Models Used

• Speaker Diarization: pyannote/speaker-diarization-3.0
• Speech Recognition: OpenAI Whisper (various sizes)
• Summarization: Custom extractive summarization algorithm

Output Files

• output.sub: Transcription in SubViewer format
• output_summary.txt: Meeting summary with key points
• output-tracks/: Directory containing audio segments for each speaker turn

License

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

Acknowledgments

• OpenAI Whisper (https://github.com/openai/whisper)
• Pyannote Audio (https://github.com/pyannote/pyannote-audio)
• Gradio (https://www.gradio.app/)
• PyTube (https://github.com/pytube/pytube)