Meta has released an open source application that creates podcasts based on uploaded documents. This could be very just like Google’s popular artificial intelligence (AI) app ‘Notebook LM’.
Meta is an app that turns PDF documents into podcasts on the twenty seventh (local time).‘NotebookLlama’It was reported that it was released as open source.
Like Notebook LM, Notebookrama converts PDF documents right into a podcast that summarizes PDF documents using Meta’s ‘Lama-3.2’ and introduces two virtual AI hosts to interactively explain the content.
First, save the PDF file of the news article or blog post as a text file using the ‘Rama-3.2-1B-Instruction’ model, after which create a podcast script from the text using the ‘Rama-3.1-70B-Instruction’ model. Do it.
Finally, the script is refined to be like a drama within the ‘Rama-3.1-8B-Instruction’ model, after which a podcast is created through the ‘parler-tts’ text-to-voice conversion model.
Nonetheless, it is understood that its performance isn’t pretty much as good as Google. The result was that the voices gave the impression of robots and overlapped one another.
The researchers found that using powerful models can improve quality. “Text-to-speech models have limits to how natural they sound,” he said.
He went on to clarify, “Performance might be improved through the use of two agents, or two models, to every create podcast summaries. Currently, a single model is used to create podcast summaries.”
Meanwhile, with the news that Google’s Notebook LM is gaining popularity, it’s predicted that other corporations will release similar apps like Meta.
Earlier this month, Gabriel Chua, an information scientist at Singapore’s government technology institute, released ‘Open NotebookLM’, which is analogous to Google’s Notebook LM, as open source. The app is built using Meta’s Lamar 3.1 405B model and MeloTTS for speech synthesis, converting PDF documents into virtual podcasts.
Reporter Park Chan cpark@aitimes.com
