From Prediction to Collaboration: Interactive Symbolic Music Analysis
arXiv cs.AI 18 hours ago
Researchers developed a unified framework for symbolic Roman-numeral music analysis that supports both high-accuracy prediction and interactive workflows including partial completion and local correction. The method was evaluated on Dilemmadata, the largest benchmark for this task, and maintains strong performance while enabling real-time iterative refinement through cached pretrained representations. The approach transforms music analysis from a pure prediction task into an interactive tool supporting targeted revision of existing labels and inference of missing annotations.