Open NotebookLM

Open NotebookLM is an open-source implementation of Google NotebookLM's popular Audio Overview feature, created by Gabriel Chua. It processes the content of a PDF or any text source, generates a natural dialogue suitable for an audio podcast, and outputs it as an MP3 file — all using open-source LLMs and text-to-speech models.

Unlike the original NotebookLM which requires Google's proprietary infrastructure, Open NotebookLM runs on your own hardware using open-source alternatives. The project is Apache 2.0 licensed and hosted at github.com/gabrielchua/open-notebooklm. It supports 13 languages and is built with a Gradio web interface for easy interaction.

Open NotebookLM requires Python 3.10+ and can be installed from source.

Source Installation

git clone https://github.com/gabrielchua/open-notebooklm.git
cd open-notebooklm
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

API Key Configuration

The project uses Fireworks AI for LLM inference (Llama 3.3 70B with JSON mode). Set your API key as an environment variable:

export FIREWORKS_API_KEY="your-api-key-here"

• PDF to Podcast — Upload a PDF and convert its content into an engaging dialogue between two AI hosts
• Web Content Support — Process web pages via URL using Jina Reader for content extraction
• Natural Dialogue Generation — Uses Llama 3.3 70B to generate informative and entertaining conversations
• Multi-language Audio — Available in 13 languages including English, Chinese, Spanish, French, and German
• Two AI Hosts — Distinct voices with natural back-and-forth interaction
• User-friendly Interface — Simple Gradio web UI for easy interaction
• MP3 Output — Downloadable high-quality audio files

Using Open NotebookLM is straightforward:

1. Launch the app: Run python app.py to start the Gradio interface
2. Upload content: Upload a PDF document or paste a URL
3. Select language: Choose from 13 supported languages for the podcast
4. Generate: Click the button to start the conversion process
5. Download: The output is delivered as an MP3 file containing the podcast dialogue

The dialogue generation uses a structured approach: the LLM first analyzes the source content, then generates a script with alternating turns between two hosts, ensuring the conversation is both informative and natural-sounding. The script is then passed to the TTS pipeline for audio generation.

Open NotebookLM combines several open-source models and tools to create the podcast pipeline:

Component	Technology	Role
LLM	Llama 3.3 70B via Fireworks AI	Dialogue script generation with JSON mode
Structured Output	Instructor library	Pydantic-based structured response parsing
TTS (Primary)	MeloTTS	English text-to-speech synthesis
TTS (Alternative)	Bark (Suno)	Alternative TTS with more expressive voices
Content Extraction	Jina Reader API	Web page content extraction from URLs
Web Framework	Gradio	Interactive web UI

Open NotebookLM can be deployed in several ways. The default Gradio app runs locally and opens in your browser. For production or shared access, it can be deployed to Hugging Face Spaces with a single click. The project includes a pre-configured app.py that works with the Hugging Face Spaces SDK.

The Gradio interface supports sharing via temporary public links (using share=True), making it easy to share with collaborators without setting up a server. The pipeline also supports custom model providers — you can swap Fireworks AI for other providers that support structured JSON output.