Open PodcastLM

Overview

Open-PodcastLM is inspired by the NotebookLM and NotebookLlama. It transforms PDF documents into engaging podcast-style conversations using opensource language models and text-to-speech technology. The tool processes PDF content, generates natural dialogues, and creates high-quality audio output featuring two distinct voices.

Built with:

• Meta LLaMA 3.1 8B, 405B via Nebius AI Studio
• ParlerTTS for Host Voice
• Bark for Guest Voice

Features

• Intelligent PDF Processing: Advanced text extraction and cleaning
• Natural Dialogue Generation: Creates engaging conversations between host and guest
• Dual Voice System: Distinct voices for host and guest using state-of-the-art TTS models
• High-Quality Audio: Professional-grade audio output with natural speech patterns

Installation

1. Clone the repository:

git clone https://github.com/krishnaadithya/open-podcastlm.git
cd open-podcastlm

2. Install dependencies:

pip install -r requirements.txt

3. Set up your Nebius API key:

export NEBIUS_API_KEY='your_api_key_here'

Command Line Arguments --pdf, -p: Path to the input PDF file (required) --output, -o: Output audio file path (default: output.mp3)

Usage

python main.py --pdf path/to/document.pdf --output podcast.mp3

Result:

Listen to Sample Generated Podcast

Project Structure

├── src/
│   ├── processors/
│   │   ├── text_processor.py
│   │   └── pdf_processor.py
│   ├── generators/
│   │   └── audio_generator.py
│   ├── clients/
│   │   └── llm_client.py
│   └── main.py
├── assets/
├── tmp/
├── README.md
└── requirements.txt

Components

• PDFProcessor: Handles PDF text extraction
• TextProcessor: Cleans and formats extracted text
• LLMClient: Manages API interactions with LLaMA models
• AudioGenerator: Generates podcast audio using dual TTS engines

Configuration

The system uses two different TTS models:

• ParlerTTS for Speaker 1 (Main host)
• Bark for Speaker 2 (Guest)

Requirements

• CUDA-compatible GPU with 24GB VRAM
• Nebius API access

License

MIT License