Reproduce “SPDMX: Large-Scale Open Stem Audio From Symbolic Music” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Phillip Long, Eduardo Escoto, Julian McAuley, Zachary Novack
SPDMX renders public-domain PDMX scores into openly licensed audio with aligned instrument stems using category-specific FluidSynth and MIDI-DDSP recipes. It releases 253,982 songs, including 88,403 multitrack songs totaling 2,734 song-hours, versus Slakh2100’s 145 hours. An author-reported, single-run Hybrid Demucs test scored 9.5 dB SI-SDR versus 0.2 dB on a 32-song SPDMX test set, an in-domain check only.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.