Reproduce “What Do Audio-Language Judges of Generated Music Listen For? Four Probes with Controls from Expert Ratings” 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
Mateusz Modrzejewski, Tomasz Radzikowski
The authors propose four controlled probes to test what influences music-judge scores: removing audio, swapping clips generated from the same prompt, reversing pair order, and rewording instructions. In tests of five open judges, all five followed the played clip for prompt alignment, but only two did so for musical quality; these findings do not establish behavior across all judges.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.