ytdebunk

Overview

ytdebunk is a command-line tool that can be installed via pip. It takes a youtube video link as an argument and does a lots of works for you automatically. This repository contains the source code and a demonstration of its features.

Current Features:

• Download audio from YouTube videos.
• Transcribe the audio content.
• Optionally enhance the transcription using the Gemini API.
• Optionally detect logical faults in the transcription using the Gemini API.
• Store the audio, transcription and logical errors in local folder

There is also a Streamlit-based demo application.

Upcoming Features:

• Classifying assertive claims within the transcription.
• Fact-checking and validating claims using online searches and agentic AI.
• Categorizing factual and logical faults.
• Generating a script for a hypothetical debunker character using generative AI (or AI Agents).
• Synthesizing the script into audio and video using generative AI (or AI Agents).

This tool is particularly useful for analyzing transcriptions to identify logical fallacies and incorrect claims made by YouTubers, helping to prepare debunk videos.

Installation

To avoid conflicts, it is recommended to create a virtual environment:

python3.11 -m venv .venv
source .venv/bin/activate

Now, install ytdebunk from PyPI:

pip install ytdebunk

Alternatively, install the latest version directly from GitHub:

pip install git+https://github.com/hissain/youtuber-debunked.git

Usage (CLI Tool)

ytdebunk is a command-line interface (CLI) with multiple options.

Arguments

• video_url (str) – URL of the YouTube video to extract audio from.

Options

Option	Description
-l, --language (str)	Language code for transcription. Supported: [bn, en] (default: en)
-e, --enhance (bool)	Enhance the transcription using the Gemini API (default: False)
-d, --detect (bool)	Detect logical faults using the Gemini API (default: False)
-v, --verbose (bool)	Enable verbose logging.
-t, --token (str)	API token for Gemini API (Required if --enhance or --detect is enabled)
-st, --start_time (float)	Start time of the audio clip (seconds)
-et, --end_time (float)	End time of the audio clip (seconds)
-m, --model (str)	Transcription model from Hugging Face (WhisperFeatureExtractor)

Example Usage

ytdebunk "https://www.youtube.com/watch?v=example" -e -d -v -t YOUR_GEMINI_API_KEY

Alternatively, using an environment variable:

export GEMINI_API_KEY="your_api_key"
ytdebunk "https://www.youtube.com/watch?v=example" -e -d -v

For more examples, check the Example Notebook.

Usage (Streamlit App)

To run the demo using Streamlit:

1. Install Streamlit:

pip install streamlit

2. Run the application:

streamlit run app.py

Screenshots of the Streamlit App

Environment Variables

Set the Gemini API token as an environment variable:

export GEMINI_API_KEY="your_api_key"

Detailed Process

1. Download Audio

   • Uses ytdebunk.downloader.download_audio to download audio from the given YouTube URL.

2. Transcribe Audio

   • Uses ytdebunk.transcriber.transcribe_audio to generate a text transcription.

3. Enhance Transcription (Optional)

   • If --enhance is enabled, ytdebunk.refiner.enhance_transcription refines the transcription using the Gemini API.
   • The API token must be provided via --token or as an environment variable.

4. Detect Logical Faults (Optional)

   • If --detect is enabled, ytdebunk.philosopher.detect_logical_faults identifies logical faults, fallacies, biases, irony, etc., using the Gemini API.
   • The API token must be provided via --token or as an environment variable.

5. Save Transcription

   • The final audio, transcription, and detected logical faults (raw or enhanced) are saved to the ./output folder.

Error Handling

• If --enhance or --detect are enabled but no Gemini API token is provided, the script exits with an error message.

License

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

Contribution & Contact

Contributions are welcome! Fork the project and submit a pull request to add new features or improve existing ones.

For inquiries, contact the project author at [email protected].