Free Music Dataset Could Make AI Music Tools Better and Cheaper
This huge free music library could lead to smarter music apps for everyone.
Researchers have introduced SPDMX, a listening-guided rendering of public-domain PDMX scores into an openly-licensed stem corpus with aligned scores and an open render pipeline. According to the article, it contains 253,982 songs, 490,159 mono stems, and 5,469 song-hours at 44.1 kHz under CC BY 4.0, with paired MIDI and identifiers linking to PDMX metadata.
Of these, 88,403 songs (34.8%; 2,734 song-hours) provide multitrack supervision. To the authors’ knowledge, this is the largest openly-licensed synthetic multitrack subset by song-hours, enabling source separation, stem-conditioned synthesis, MIDI-to-audio, transcription, and joint symbolic-audio modeling. The corpus, metadata, and open pipeline are released for reuse on other symbolic collections, with demos and experimental notes on the project page.
- SPDMX is a free dataset of over 253,000 songs with separate instrument tracks, created from public-domain music scores.
- It's the largest openly licensed dataset of its kind, enabling AI research without copyright worries.
- This could lead to better music apps for consumers, like improved vocal removal or instrument isolation tools.
Why It Matters
Better AI music tools could give you more control over songs for fun, learning, or creative projects.