Skip to content
IFAI

Open source

Meta open-sources a clone of Google's viral podcast maker

Published Reach 45

Meta just dropped NotebookLlama, an open-source project that recreates Google NotebookLM's viral podcast generator using its own Llama models. The pipeline turns uploaded files into dramatic dialogue scripts before piping them into open text-to-speech tools. Why it matters: The audio still sounds robotic and cuts each other off, but developers now have an open recipe to customize and upgrade.

Meta open-sources a clone of Google's viral podcast maker
Image: techcrunch.com

NotebookLlama is an open-source tool from Meta that turns uploaded text files into a back-and-forth, podcast-style audio digest. Basically, it is an open recipe designed to mimic the viral Audio Overview feature in Google's NotebookLM.

How it turns text into a show

Unsurprisingly, the pipeline relies heavily on Meta's own Llama models.

You start by feeding it a text file, such as a PDF of a news article or a blog post. NotebookLlama first creates a clean transcript of the source material. Next, the model rewrites that text to add dramatization and conversational interruptions, generating a dialogue script between two hosts.

That final script is then funneled into open-source text-to-speech models to spit out the generated audio.

On the project's GitHub page, Meta researchers noted that a single model is currently doing all the heavy lifting to write the podcast outline. They pointed out another possible route for future updates: having two distinct AI agents debate the topic of interest and craft the outline together.

What it sounds like right now

Do not expect this to go toe-to-toe with Google NotebookLM straight out of the box.

In the audio samples released so far, the hosts suffer from a very obvious robotic quality. On top of that, the voices have an awkward habit of talking over each other at odd points in the conversation, making the banter feel unnatural.

The researchers behind the project openly acknowledged this bottleneck. They wrote that current open text-to-speech models are the main limitation on how natural the audio sounds, though stronger models should eventually help bridge the gap.

There is also the persistent issue of AI hallucinations. Just like NotebookLM itself and other community clones, NotebookLlama cannot guarantee 100% accuracy, meaning the synthesized hosts will inevitably make things up that were never in your original document.

Why it matters

Google's NotebookLM blew up because the banter felt eerily human, but it remains a proprietary walled garden. NotebookLlama flips that dynamic on its head by giving everyone the complete blueprint.

If you just want a polished podcast from your meeting notes right now, stick with Google. But for developers and tinkerers, NotebookLlama is an open sandbox. You can host it yourself, hook up better open-source speech engines, or experiment with multi-agent debate setups without paying platform fees or handing private files over to Google.

Sources (1)

  1. [1]
    techcrunch.com(opens in a new tab)Originaltechcrunch.com

Best of AI, daily