NotebookLM podcast not being identified

[johnspurlock]

Here's a podcast that uses NotebookLM, but python is telling me: The audio is: Human for the first episode mp3

https://serve.podhome.fm/rss/2cd114fe-ae01-4d1e-90b0-22f7e51618b9

Can you confirm? Want to rule out some python setup issue on my side.

[wenbinf]

Hi @johnspurlock

Confirmed. The current model is unable to correctly classify this audio as AI-generated.

It's likely that the audio has been edited, differing from other raw NotebookLM-generated audio files.

[wenbinf]

Hi @johnspurlock

Confirmed. The current model is unable to correctly classify this audio as AI-generated.

It's likely that the audio has been edited, differing from other raw NotebookLM-generated audio files.

I just trained on more audio files (a few hundreds), and the latest model works for this specific audio.

You can git pull & try again.

[johnspurlock]

et voilà! The audio is: AI Generated

What is the best way to help this effort? Send PRs to the CSV when new notebooklm shows are found?