Build a music dataset from
eleven open sources.
Song metadata, time-synced lyrics and full-length audio — collected into one flat folder and exported to Excel, SQL, PostgreSQL and JSON. No API keys, nothing to sign up for.
- Sources
- 11
- Seed artists
- 926
- API keys required
- 0
Console
Set the size of the run, choose what goes in it, then start.
- Local
- 0
- Intl
- 0
- Full audio
- 0
- Skipped
- 0
- Failed
- 0
Every step — sources queried, tracks saved, failures — shows up here once a run starts.
Collected tracks
Everything the current run has written to disk.
| # | Artist | Title | Year | Region | Audio | Lyrics |
|---|---|---|---|---|---|---|
| No tracks yet — start a run in the console above. | ||||||
How a run works
Four stages, in order. Each one hands its output to the next.
-
1
Discover
Chart and metadata sources are queried for the seed artists in the genres you picked. Slow sources are skipped once there are enough candidates.
-
2
Filter
Candidates are merged across sources, deduplicated four ways — normalised key, ISRC, fuzzy title, then audio hash — and balanced between local and international.
-
3
Fetch
Full-length audio via yt-dlp, time-synced lyrics from LRCLIB, and release metadata backfilled from Wikipedia. Every file is verified before it counts.
-
4
Export
One flat output folder plus combined exports: Excel, generic SQL, a PostgreSQL dump and JSON.
Sources
Every one is open. Anything that demanded a key, token or OAuth was removed from the project rather than left switched off.