Reproduce “Training Music Sample Identification Models on Real Sample Pairs” 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
R. Oguz Araz, Joan Serrà, Xavier Lizarraga-Seijas, Xavier Serra, Yuki Mitsufuji, Dmitry Bogdanov
SIE trains embeddings on real annotated source-sampler pairs, varying segments in pitch, timing, and other ways. On WhoSampled130K’s 10,444-query test, mean average precision (mAP), summarizing relevant-track rankings, was 0.389 ± 0.009, versus 0.299 ± 0.008 for SampleID-R and 0.226 ± 0.007 for SampleID. With 5% of the annotated pairs, SIE matched SampleID; real data alone does not explain its gains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.