Meta has unveiled a new tool called “NotebookLlama,” a feature similar to Google’s NotebookLM, which creates podcast-like content from text files.

NotebookLlama relies on Meta’s Llama models for most of its processing. Like NotebookLM, it generates conversations in a podcast format from uploaded files.

Here’s how it works: NotebookLlama first transcribes a file, such as a news article or blog post in PDF form. It then adds elements like dramatization and interruptions, before converting the transcript into speech using open-source text-to-speech models.

Buy Me A Coffee

The quality, however, isn’t quite on par with NotebookLM. From the samples I’ve heard, the voices sound noticeably robotic, and at times, they overlap awkwardly.

Meta’s researchers acknowledge this issue, stating that better models could enhance the audio quality. They also mentioned on the project’s GitHub page that the current limitation comes from the text-to-speech model. They suggest another approach could involve two AI agents debating a topic to generate a podcast outline, whereas right now, they use just one model for this.

This isn’t the first attempt to mimic NotebookLM’s podcasting feature, and while some have been more successful than others, none—including NotebookLM—have fully overcome the problem of AI-generated “hallucinations” where the content might include made-up information.

READ
AdSense Inbox Notifications to Be Deprecated in November 2024