mir-final-project

Classification of Mozart and Beethoven scores with extracted features such as pitch histogram and note density using music21.

Dataset retrieved from https://github.com/musescore/MuseScore/tree/30557dec4e573f0aa5e78043da1b7564da222452

Metadata file score.jsonl retrieved from https://musescore-dataset.xmader.com/score.jsonl

Usage

process_data.py

Using the metadata file, we can retrieve information about instruments and the composers in the description.

To retrieve the piano only scores, use the --process, which will save the list of piano-only MuseScore files from Mozart and Beethoven in the data folder.

To convert the .mscz files into MusicXML (.musicxml), you can run the script with --convert which will require the pickle files generated with --process option. It will create the .musicxml files in the same directories as the .mscz files. Runs the MuseScore program as a subprocess (mscore) to convert all the files.

WARNING: the --convert option might get stuck for unknown reasons, you can run musescore.mscore -j data.json manually to convert your data. With data.json being the json containing the input in .mscz and output in .musicxml. You can use the function create_convert_batch to create your json file. For more information: https://musescore.org/en/handbook/3/command-line-options#Run_a_batch_job_converting_multiple_documents

The script will generated pickle files that contain the list of Mozart and Beethoven in MusicXML format.

python process_data.py <musescore_directory> --process --convert

After getting the list of Mozart and Beethoven files in MusicXML format, you will find the separated .mscz and .musicxml datasets in dataset_mscz and dataset_musicxml.

extract_features.py

Extract features from Mozart and Beethoven MusicXML files such as pitch histogram and note density (or complexity) in csv files. Requires dataset_musicxml with subfolders mozart and beethoven containing MusicXML files.

classification.py

Uses Random Forests to classify between Mozart and Beethoven scores and computes feature importance.