README.md

metadata

pretty_name: MuSP-Bench
language:
  - en
task_categories:
  - question-answering
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/questions.csv

MuSP-Bench

MuSP-Bench is a 456-question benchmark for musical score understanding, performance listening, and combined score-performance reasoning.

Contents

• data/questions.csv: all 456 questions and the response contract for each.
• scores/cleaned_reorganized/: one cleaned and reorganized PDF per piece.
• scores/without_context/: one context-removed PDF per piece.
• audio/: one shared MP3 performance per piece.

Each question row points to its two score PDFs and audio file. Media files are reused by every question about the same piece.

Modalities

• S: answer from the score.
• P: answer from the performance recording.
• SP: answer using both score and performance.
• S/P: answer from either source.

answerExample contains JSON-encoded formatting examples. Examples illustrate the requested syntax only and are not answers to the benchmark questions.

Ground-truth answers are intentionally not included in this release.