[ad_1]
Meta has released an “open” implementation of Google’s viral podcast creation feature NotebookLM.
Named NotebookLlamaUnsurprisingly, the project uses Meta’s Llama models for most of its processing. Like NotebookLM, it can create podcast-style synopses of text files uploaded to it.
NotebookLlama first creates text from a file – for example a PDF of a news article or blog post. It then adds “more drama” and interruptions before feeding the text to open text-to-speech forms.
The results don’t look nearly as good as NotebookLM. In the llama’s notebook Samples I listened to the voices having a very clear robotic quality, and they tended to talk over each other at odd points.
But the meta-researchers behind the project say quality can be improved with stronger models.
“The text-to-speech paradigm is limiting how natural this can be,” they wrote on NotebookLlama’s GitHub page. “(Also) another way to write a podcast is to have two agents discuss the topic of interest and write a podcast outline. We currently use one template for writing a podcast outline.
NotebookLlama is not the first attempt To replicate NotebookLM’s podcast feature. Some projects have been more successful than others. But none of them — not even NotebookLM itself — have solved the hallucination problem plaguing AI. This means that AI-generated podcasts are just as well It must contain some make-up items.
[ad_2]