Synthetic Dataset Generation for String Ensemble Separation

forked musicxml_parser from https://github.com/qsdfo/musicxml_parser
and added some more articulations

Data Creation from MusicXML

from xmlToMidi import xmlToMidi

paths = ["Vn1_path.musicxml",
         "Vn2_path.musicxml",
         "Va_path.musicxml",
         "Vc_path.musicxml"]
xmlToMidi(paths, save_path="path/to/file/without/.mid")

  REQUIRED: manually add tempo mark for splitting (Use any DAW program)

  split midi based on manual tempor marks

from splitMidiData import splitMidiData

splitMidiData("path/to/file/after/tempo")

  generate synthesis parameters before audio

from generate_dataset import generate_synth

path_format = "folder/after/split/{:02d}_{:01d}_{}.mid"
save_format = "path/to/audio/{:02d}/{:01d}.npz"

splits = 16  ## number of tempo changes
generate_synth(path_format, save_format, splits=splits)

  generate audio from synth parameters

from generate_dataset import generate_audio_from_synth

path_format = "path/to/audio/{:02d}/{:01d}.npz"

splits = 16  ## number of tempo changes
generate_audio_from_synth(path_format, splits, interpolation_rate=0.99)