Meta Launches Open Source Alternative to Google's NotebookLM Podcast Feature
Meta introduces NotebookLlama, an open-source tool that generates podcast-style audio from text documents using Llama models. The current output sounds noticeably robotic compared to Google's version, but researchers see a clear path to improvement.
Meta releases an open-source version of Google's popular NotebookLM podcast generator called NotebookLlama. The new project uses Meta's Llama models to process uploaded text files, such as PDF articles or blog posts, and converts them into two-host, conversational podcast episodes.
The current audio output lacks the polish of Google's version, as the voices exhibit a distinctly robotic quality and occasionally talk over each other. Meta researchers admit that the underlying text-to-speech model limits the natural sound of the audio and note that using a single model to write the outline restricts the conversational dynamic.
NotebookLlama joins a growing list of attempts to replicate NotebookLM's viral feature, though none of these tools completely solve the persistent AI hallucination problem. Because the technology relies on generative AI to script the dialogue, the resulting audio discussions inevitably contain fabricated information alongside the actual source material.