Meta has launched an “open” implementation of the viral generate-a-podcast function in Google’s NotebookLM.
Known as NotebookLlama, the challenge makes use of Meta’s personal Llama fashions for far of the processing, unsurprisingly. Like NotebookLM, it could possibly generate back-and-forth, podcast-style digests of textual content information uploaded to it.
NotebookLlama first creates a transcript from a document — e.g. a PDF of a information article or weblog publish. Then, it provides “extra dramatization” and interruptions sooner than feeding the transcript to open text-to-speech fashions.
Symbol Credit:Meta
The consequences don’t sound just about as excellent as NotebookLM. Within the NotebookLlama samples I’ve listened to, the voices have an overly clearly robot high quality to them, and have a tendency to speak over each and every different at strange issues.
However the Meta researchers at the back of the challenge say that the standard might be progressed with more potent fashions.
“The text-to-speech fashion is the limitation of ways herbal this may occasionally sound,” they wrote on NotebookLlama’s GitHub web page. “[Also,] some other way of writing the podcast can be having two brokers debate the subject of hobby and write the podcast define. Presently we use a unmarried fashion to put in writing the podcast define.”
NotebookLlama isn’t the primary try to reflect NotebookLM’s podcast function. Some tasks have had extra luck than others. However none — now not even NotebookLM itself — have controlled to unravel the hallucination downside that canine all AI. This is to mention, AI-generated podcasts are certain to comprise some made-up stuff.