FallacyScoreboard

FallacyScoreboard is an offline tool that downloads YouTube videos, processes them using WhisperX for automatic speech recognition and speaker diarization, detects logical fallacies using LLMs with OLLAMA, and overlays the results on the video with a scoreboard.

Demo Video

Click the image above to watch a demonstration of FallacyScoreboard in action!

Features

Download YouTube videos and extract audio
Transcribe audio using WhisperX with word-level timestamps and speaker diarization
Detect logical fallacies in transcribed text
Overlay fallacy detection results on the video with a scoreboard and side panel
Handle multiple fallacies per speaker segment
Color-code speakers for easy identification

Installation

Clone the repository:

git clone https://github.com/latent-variable/FallacyScoreboard.git
cd FallacyScoreboard

Create a Python environment:

conda create --name fallacyscoreboard python=3.10
conda activate fallacyscoreboard

Install dependencies:
```
pip install -r requirements.txt
```
Install ImageMagick: FallacyScoreboard uses MoviePy, which requires ImageMagick. Make sure to install ImageMagick and set the path in the script:
```
change_settings({"IMAGEMAGICK_BINARY": r"path/to/your/ImageMagick-7.1.1-Q16-HDRI/magick.exe"})
```

Usage

Process a YouTube video:
```
from fallacyscoreboard import fallacy_detection_pipeline

output_path = "path/to/output/directory"
video_url = "https://www.youtube.com/watch?v=example"

fallacy_detection_pipeline(video_url, output_path)
```
This function will:
- Download the YouTube video
- Extract audio from the video
- Transcribe the audio using WhisperX
- Detect fallacies in the transcription
- Overlay the fallacies and scoreboard on the video
Customize OLLAMA settings: You can customize the OLLAMA host and model by modifying these variables in the script:
```
OLLAMA_HOST = 'http://localhost:11434'
OLLAMA_MODEL = 'llama3.1'
```

Output

The script will generate several files in the output directory:

The downloaded video file
An extracted audio file
A JSON file with the transcription and diarization results
A text file with formatted transcription
A JSON file with fallacy analysis results
The final video with overlaid fallacy scoreboard and explanations

License

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

Notes

Ensure you have sufficient disk space and computational resources, as processing videos can be resource-intensive.
The quality of fallacy detection depends on the OLLAMA model used. Experiment with different models for best results.
This tool is for educational and research purposes. Always respect copyright and terms of service when using YouTube content.