Fake Conversations Generator

A cybersecurity demonstration tool that shows how audio can be manipulated to create misleading content by selectively extracting and recombining segments from original recordings.

Created on Sundai, April 6, 2025.

⚠️ Educational Purpose Only

This tool is designed strictly for educational and cybersecurity awareness purposes. It demonstrates how easily media can be manipulated to create false narratives, highlighting the importance of media literacy and verification in the digital age.

Overview

The Fake Conversations Generator automates the process of:

1. Downloading audio from YouTube videos
2. Transcribing the audio with word-level timing information
3. Analyzing the transcript to identify segments that could be recombined to create misleading content
4. Extracting and combining those segments to create a fake audio clip

Features

• YouTube Integration: Download audio and transcripts directly from YouTube videos
• High-Quality Transcription: Uses ElevenLabs Speech-to-Text API for accurate word-level transcription with timing
• AI-Powered Analysis: Leverages Google's Gemini AI to identify segments that could be repurposed
• Audio Manipulation: Extracts and combines audio segments using FFmpeg

Requirements

• Python 3.8+
• FFmpeg
• API keys:
  • ElevenLabs API key
  • Google Gemini API key

Installation

1. Clone this repository

2. Install required Python packages:

   pip install -r requirements.txt
   

3. Install FFmpeg (if not already installed):

   • macOS: brew install ffmpeg
   • Ubuntu: sudo apt install ffmpeg
   • Windows: Download from ffmpeg.org

4. Create a .env file in the project root with your API keys:

   ELEVENLABS_API_KEY=your_elevenlabs_api_key
   GEMINI_API_KEY=your_gemini_api_key
   

Usage

Run the main script with a YouTube video ID:

./do.sh <youtube_video_id>

For example:

./do.sh dQw4w9WgXcQ

Step-by-Step Process

1. Download: The script downloads the audio from the specified YouTube video
2. Transcribe: The audio is transcribed using ElevenLabs API with word-level timing
3. Analyze: Gemini AI analyzes the transcript to identify segments that could be recombined
4. Create: The identified segments are extracted and combined to create a fake audio clip

Individual Components

You can also run each component separately:

YouTube Downloader

python youtube_downloader.py "https://www.youtube.com/watch?v=<video_id>"

ElevenLabs Transcriber

python elevenlabs_transcriber.py "downloads/<video_id>.mp3"

Gemini Analyzer

python gemini_analyzer.py "downloads/<video_id>_elevenlabs_transcript.json"

Fake Audio Creator

python create_fake_audio.py "downloads/<video_id>_elevenlabs_transcript_compromising_with_timing.json" "downloads/<video_id>.mp3"

Output Files

The process generates several files in the downloads directory:

• <video_id>.mp3: Original audio from YouTube
• <video_id>_transcript.json & <video_id>_transcript.txt: YouTube's transcript (if available)
• <video_id>_elevenlabs_transcript.json & <video_id>_elevenlabs_transcript.txt: ElevenLabs transcript with word-level timing
• <video_id>_elevenlabs_transcript_compromising.json: Segments identified by Gemini as potentially misleading
• <video_id>_elevenlabs_transcript_compromising_with_timing.json: Same as above but with timing information
• <video_id>_elevenlabs_transcript_compromising_fake.mp3: The generated fake audio clip

Ethical Considerations

This tool demonstrates how easily audio can be manipulated to create misleading content. It is intended to:

• Raise awareness about the ease of creating fake content
• Encourage critical thinking when consuming media
• Promote the development of detection technologies

Do not use this tool to create misleading content for malicious purposes.

License

This project is provided for educational purposes only.