Reproduce “SMEBENCH: Benchmarking Neural Music Editing with Synthetized MIDI Edits” 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
Matthieu Cervera, Francesco Paissan, Micael Antunes, Mirco Ravanelli, Ricard Marxer, Cem Subakan
The authors introduce SMEBench, a benchmark for evaluating music edits against paired references. They apply controlled changes to symbolic performances and render matched source and edited audio, separating preservation, target agreement, and text compatibility. Across six systems under edit-caption conditioning, MusicGen leads CLAP-Text at 0.350, while SteerMusic leads target CLAP-Audio and LPAPS; these metric-specific results do not establish a definitive editor ranking.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.