A curated collection of 8 short polyphonic phrases.

This dataset contains musical excerpts originally used as stimuli in behavioral experiments investigating note-level structural salience in polyphonic music. Each phrase is selected to be:

• Short: 8-16 bars, suitable for iterative reduction tasks
• Tonally coherent: Begins and ends in the same key (or returns to tonic)
• Structurally complete: Contains clear phrase boundaries
• Polyphonically rich: Features multiple voices with melodic and harmonic content

short_phrase_collections/
├── README.md
├── LICENSE
├── metadata/
│   └── phrases.csv          # Main metadata (single source of truth)
├── stimuli/
│   ├── phrase_01/           # Purcell - Rondeau from Abdelazer
│   │   ├── score.musicxml
│   │   └── score.pdf
│   ├── phrase_02/           # Handel - Suite HWV 450 Menuet
│   └── ...
└── docs/
    └── selection_criteria.md

Each note in a stimulus file is assigned a unique ID following this traversal order:

1. Sort all notes by offset (onset time) ascending
2. For notes at the same offset, sort by midi pitch descending (higher pitch first)
3. Assign sequential IDs: phrase_01_1, phrase_01_2, ...

This ensures deterministic, reproducible ID assignment across processing pipelines.

Some stimuli required preprocessing to remove duplicate notes (same onset + same pitch):

Policy: When duplicates exist, the note with the longest duration is kept. Full details are recorded in metadata/phrases.csv.

from music21 import converter

score = converter.parse('stimuli/phrase_01/score.musicxml')
for note in score.flatten().notes:
    print(note.pitch, note.offset, note.quarterLength)

import pandas as pd

phrases = pd.read_csv('metadata/phrases.csv')
print(phrases[['phrase_id', 'composer', 'work_title', 'key']])

See docs/selection_criteria.md for detailed inclusion/exclusion criteria.