pyScoreParser

musicxml & performance midi alignment managing tool.

this repository can be used for:

• parse score xml into a note sequence while preserving score information.
• match midi notes in performance midi to score note, and make combined representation.

Installation

git pull https://github.com/jdasam/pyScoreParser.git
# install midi_utils, musicxml_parser
git clone https://github.com/mac-marg-pianist/musicXML_parser.git
cd musicXML_parser
pip install .
cd ..

git clone https://github.com/mac-marg-pianist/midi_utils.git
cd midi_utils
pip install .
cd ..

# install AlignmentTool from https://midialignment.github.io/demo.html
# and compile
wget https://midialignment.github.io/AlignmentTool_v220127.zip
unzip AlignmentTool_v220127.zip
cd AlignmentTool
./compile.sh
cd ..

usage

# load piece, which contains score and performances

from pyScoreParser.data_class import PieceData


score_xml = 'test_examples/Beethoven/32-1/musicxml_cleaned.musicxml'

piece = PieceData(xml_path)

test_examples/Beethoven/32-1/musicxml_cleaned_score.mid

Number of mismatched notes: 43

Performance path is test_examples/Beethoven/32-1/DupreeF03.mid

Number of Matched Notes: 5296, unmatched notes: 67

print(piece.__dict__.keys())

dict_keys(['meta', 'performances', 'score', 'xml_obj', 'xml_notes', 'num_notes', 'notes_graph', 'score_midi_notes', 'score_match_list', 'score_pairs', 'measure_positions', 'beat_positions', 'section_positions', 'score_features'])

# check first note in score

print(len(piece.score.xml_notes))
print(piece.score.xml_notes[0])

5363

{duration: 60, midi_ticks: 27.5, seconds: 0.20833333333333334, pitch: Eb4, MIDI pitch: 63, voice: 5, velocity: 64} (@time: 0.0) (@xml: 0)

# check first note in performance
print(len(piece.performances[0].midi_notes))
print(piece.performances[0].midi_notes[0])

5531

Note(start=0.504807, end=0.560897, pitch=63, velocity=91)

# check alignment
print(len(piece.performances[0].match_between_xml_perf))
print(piece.performances[0].match_between_xml_perf[:10])

5363

[0, 1, 2, 3, 4, 7, 6, 5, 8, 9]a

load dataset

checkout test_dataset.py

use yamaha dataset or emotion dataset.

if save, save ScoreData and PerformData into .dat format.

if not save, load from pre-saved file