Reproduce “MIDI-REPA: Improving Text-to-Music Generation with Symbolic Representation Alignment” 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

Song-Ze Yu, Fang-Duo Tsai, Yi-Hsuan Yang

MIDI-REPA trains a text-to-music model to align hidden representations with automatic music transcription targets. Using a MuScriptor-derived piano roll at layer L2, the 122-million-parameter model reports FAD 0.497 and CLAP similarity 0.315, versus baseline scores of 0.757 and 0.088 on 80 ATTM prompts. These objective results concern short-form generation and do not establish performance on full songs or larger models.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.