Meta NotebookLlama AI Podcast Generator Launched as ‘Open Supply Instrument’ to Tackle Google’s NotebookLM

Meta launched a brand new open-source synthetic intelligence (AI) instrument on Sunday that may tackle the Google NotebookLM. Dubbed NotebookLlama, the instrument is an AI-powered podcast generator the place customers can add a PDF file and the instrument will flip it into an audio podcast with two AI characters. The instrument makes use of three totally different Llama 3.1 AI fashions to finish your complete course of. Identical to Google’s instrument, NotebookLlama’s podcast additionally follows a back-and-forth dialog between two AI hosts in a free-flowing method.

The Meta NotebookLlama AI instrument makes use of three giant language fashions to generate audio podcasts from blocks of textual content. At the moment, the instrument solely accepts PDF recordsdata as enter, so customers must convert no matter textual content format they’ve into PDF.

Meta NotebookLlama workflow
Photograph Credit score: Meta

NotebookLlama first makes use of Llama 3.2 1B instruct mannequin to pre-process the PDF file and reserve it in a ‘.txt’ file. Then the Llama 3.1 70B instruct mannequin is used to write down a podcast transcript utilizing the supply dataset. The transcription is then dramatised utilizing a re-writer which makes use of the Llama 3.1 8B instruct mannequin. Lastly, a customized instrument is used so as to add the transcription in a text-to-speech workflow. For this, Meta is utilizing the Parler TTS instrument. people can entry all of the fashions required to generate podcasts from the GitHub itemizing right here.

Nevertheless, the AI fashions talked about above are simply suggestions from the builders. Customers can want to make use of smaller fashions for each step, nevertheless, the outcomes might fluctuate. Meta highlighted that to run the AI system within the advisable setup, customers would require a GPU with an aggregated reminiscence of roughly 140GB.

An X (previously often known as Twitter) person posted a pattern of the generated podcast. Based mostly on this, it seems the audio high quality is inferior to the Google NotebookLM, and it sounds shrill and robotic. Additional, there are situations the place components of audio get ignored and the AI hosts find yourself talking over one another.

Meta acknowledges among the points and plans to enhance them within the subsequent iteration of the AI product. The corporate highlighted, “The TTS mannequin is the limitation of how pure this may sound. This in all probability be improved with a greater pipeline and with the assistance of somebody extra educated.”

The tech large can be planning to make use of two totally different LLMs to write down the script, the place every mannequin will debate the opposite to make the podcast sound extra conversational. That is additionally a part of the builders’ future pipeline. Moreover, the corporate can be testing the Llama 405B AI mannequin to write down the transcripts in addition to rising the help for extra enter and output codecs.