Datasets:
language:
- en
- es
- de
- ja
- zh
- nl
- 'no'
- fr
- uk
- sv
- it
- da
- fi
- pt
license: other
license_name: public-domain
license_link: LICENSE
task_categories:
- text-to-speech
- automatic-speech-recognition
size_categories:
- 1K<n<10K
tags:
- audio
- webdataset
- synthetic-speech
- notebooklm
- multilingual
pretty_name: NotebookLM Podcasts
configs:
- config_name: default
data_files:
- split: train
path: '*.tar'
NotebookLM Podcasts
A multilingual collection of 8,223 podcast episodes totaling 2,411.5 hours, organized as WebDataset TAR shards. Google NotebookLM is a research and note-taking tool that can generate podcast-style audio discussions from source material. The collection contains AI-generated audio; some episodes also contain human introductions or editing.
License: Public Domain.
Collection
An Agentify-style search used parallel agents to find publicly available podcasts, RSS feeds, and direct audio links. Original feed and episode metadata were retained with the audio. NotebookLM attribution and spoken language follow the available publisher and feed information; not every recording has been independently listened to.
Listen to examples
Two complete episodes per language. Sign in to access the audio in this private dataset.
English
Ep.59/1803 - Ontological Dependence — ChatSEP (7:34)
Ep.45/1803 - Medieval Theories: Properties of Terms — ChatSEP (7:41)
German
OX.ai 18/18 - Robert Sapolsky: Individuelle Unterschiede und die wichtigste Aussicht — Wissen in Bewegung (6:14)
OX.ai 14/18 - Robert Sapolsky: Menschliches Sexualverhalten II: Pheromone und Abwechslung — Wissen in Bewegung (6:33)
Spanish
El brócoli entre los donuts: Por qué el contenido serio se siente incómodo en YouTube — RGartner Audios Español (5:58)
El influencer equivocado: Por qué fracasan las campañas publicitarias — RGartner Audios Español (7:38)
Dutch
BoerBurgerBeweging – BBBeter Nederland — Politieke Bla Bla (10:47)
CDA – Bouwen op vertrouwen — Politieke Bla Bla (18:54)
French
⏪ Le récap' de Deep Media 🎙️ Mélissa Hébert, directrice plateformes à Télé-Québec — Deep Media (6:07)
⏪ Le récap' de Deep Media 🎙️ Sandrine Roustan, directrice générale des contenus de la RTBF — Deep Media (6:11)
Statistics
| Language | Episodes | Hours |
|---|---|---|
| English | 6,624 | 1,975.3 |
| Spanish | 662 | 209.4 |
| German | 319 | 96.6 |
| Japanese | 142 | 33.6 |
| Chinese | 126 | 16.7 |
| English / Dutch – unresolved | 104 | 29.5 |
| Norwegian | 66 | 8.0 |
| French | 57 | 13.6 |
| Dutch | 38 | 12.8 |
| Ukrainian | 34 | 4.1 |
| Swedish / English – unresolved | 17 | 4.3 |
| Italian | 15 | 3.8 |
| Swedish | 8 | 0.8 |
| Danish | 5 | 0.7 |
| Finnish | 4 | 1.1 |
| Portuguese | 1 | 0.5 |
| English / German – unresolved | 1 | 0.7 |
| Total | 8,223 | 2,411.5 |
Every episode has a known duration. Mixed-language labels indicate that the spoken language has not been resolved.
Dataset structure
Each TAR shard contains paired files sharing the same key: KEY.mp3, KEY.m4a, KEY.wav, or KEY.mp4, and KEY.json. Most audio is MP3 or M4A; the collection also includes WAV and one MP4 containing audio and video. File extensions follow the container signature.
The JSON metadata can include source and episode information, RSS XML or publisher-page HTML, source and media URLs, timestamps, HTTP response details, provenance notes, file size, and a SHA-256 checksum. Available fields vary by source; session-cookie header values are redacted.
shard-00000.tar
KEY.mp3
KEY.json
...