notebooklm-podcastlisted

# NotebookLM Podcast Skill Drive the NotebookLM → podcast → dataset pipeline in this repo. It uses the unofficial [`notebooklm-py`](https://github.com/teng-lin/notebooklm-py) API (pure HTTP after a one-time browser login), plus `faster-whisper` (transcription) and `pyannote` (2-speaker diarization). ## Prerequisites (check first) - `notebooklm login` has been run once (auth in `~/.notebooklm/profiles/default/storage_state.json`). If a command returns "Authentication expired", tell the user to run `notebooklm login` — it needs a browser and cannot be automated. - `ffmpeg` in PATH. For transcription/diarization: a GPU (Colab T4) — CPU works but is slow. ## Common tasks ### Generate a podcast from a topic Deep-research the topic on NotebookLM, then generate one or more audio overviews. ```bash notebooklm create "TOPIC" --json # -> notebook id (in .notebook.id) notebooklm source add-research -n <nb_id> "TOPIC" --mode deep --json notebooklm research wait -n <nb_id> --import-all --json notebooklm generate audio "custom prompt for the hosts" -n <nb_id> --length long --wait --json # generate returns {task_id, status, url} — NOT artifact_id. Get it from the list: notebooklm artifact list -n <nb_id> --json # newest audio artifact -> its id notebooklm download audio -n <nb_id> -a <artifact_id> out.mp3 ``` Other output types (same command, free): `video`, `quiz`, `mind-map`, `flashcards`, `slide-deck`, `infographic`, `data-table`, `report`. ### Recov