AI PodcastifyAI

AI PodcastifyAI is an application that transforms scientific papers and web content into engaging podcast-style conversations using artificial intelligence. This tool leverages advanced language models and text-to-speech technology to create informative and accessible audio content from complex textual information. This is poor man's version of Google's NotebookLM AI Podcast.

demo.mp4

sample-audio.mp4

Features

• Text Input: Enter scientific text or a webpage URL directly into the application.
• AI-Powered Dialogue Generation: Utilizes KoboldCPP to generate a natural conversation between two speakers based on the input content.
• Text-to-Speech Conversion: Employs StyleTTS2 to convert the generated dialogue into lifelike speech.
• Multi-Voice Support: Creates a dynamic listening experience with distinct voices for different speakers.
• Audiobook Creation: Combines individual audio segments into a cohesive MP3 audiobook.
• User-Friendly GUI: Offers an intuitive interface for easy interaction and processing.

Requirements

• Python 3.8+
• tkinter
• requests
• BeautifulSoup4
• StyleTTS2 API
• scipy
• pydub
• numpy
• tortoise-tts

Installation

1. Clone this repository:

   git clone https://github.com/PasiKoodaa/ai-podcastify.git
   cd ai-podcastify
   

2. Install the required dependencies:

   pip install -r requirements.txt
   

3. Prepare voice samples named melinda_voice.wav and steve_voice.wav for the two speakers and put them in the same folder where the file "main.py" is.

Usage

1. Ensure you have KoboldCPP running locally on port 5001.

2. Run the application:

   python main.py
   

3. In the GUI:

   • Enter the text of a scientific paper or a webpage URL in the input area.
   • Click "Process" to generate the podcast dialogue.
   • Once processing is complete, click "Create Audiobook" to generate the MP3 file.

4. The resulting audiobook will be saved as audiobook.mp3 in the same directory.

How It Works

1. Text Processing: The app fetches content from the provided text or URL.
2. Dialogue Generation: KoboldCPP generates a conversational dialogue based on the input.
3. Text-to-Speech: StyleTTS2 converts the dialogue into speech for each speaker.
4. Audio Compilation: Individual audio segments are combined into a single MP3 file.

Limitations

• Requires a local instance of KoboldCPP running on port 5001.
• Processing time may vary based on input length and system capabilities.
• Internet connection required for webpage content fetching.

Acknowledgments

• KoboldCPP for dialogue generation: https://github.com/LostRuins/koboldcpp
• StyleTTS2 API for text-to-speech conversion: https://github.com/NeuralVox/StyleTTS2 (At the moment this is really hard to get to work on Windows)
• All other open-source libraries used in this project