Meta has quietly released NotebookLlama, an open source clone of that Notebook LM podcast generator from Google. Quoting Techopedia:
Similar to Google’s NotebookLM, which helps users with a one-click podcast generation feature, Meta’s NotebookLlama enables users to turn documents into conversational audio featuring two AI voices.
NotebookLlama processes user-uploaded PDF files through a series of steps involving Meta’s Llama 3.1 models. Initially, the Llama 3.2 1B model converts the PDF into text format. The Llama 3.1 70B model then crafts a podcast script from the text, while a smaller Llama 3.1 8B model adds a conversational tone. Finally, Meta’s Parler TTS tool transforms the script into audio, producing a dynamic dialogue between the AI characters.
While many have shared their excitement and praised the tool’s potential in early feedback, some have highlighted several limitations.
Users reported on X that the audio outcome was not as smooth as the ones from NotebookLM, with occasional overlaps where the AI speakers unintentionally talk over each other.
Meta has acknowledged these issues, and in its roadmap for NotebookLlama, envisions upgrades to enhance dialogue flow and sound quality. Planned improvements include using different language models for each AI character to better simulate back-and-forth discussions in a way that makes the audio sound human-like.
#ainews