Chord-Jazzification

A deep learning approach to the chord coloring and the chord voicing. A dataset featuring the interpretations of chord symbols is also propossed.

See the paper for more information: Chord Jazzification: Learning Jazz Interpretations of Chord Symbols (ISMIR 2020)

Requirements

• tensorflow-gpu 1.8.0
  
• numpy 1.16.2
  
• pretty_midi 0.2.9
  

Descriptions

• Chord_Jazzification_Dataset: the annotations of 50 pop-jazz piano solos

• chord_jazzification_preprocessing.py: preprocess the chord jazzification dataset and get the preprocessed data chord_jazzification_training_data.pickle

  To listen to the chord progressions of the dataset, uncomment the following code:
  #generate_midi_instance(corpus['1'], 'example.mid', qpm=120, play_midi=True, show_pianoroll=False)

  You can listen to other pieces of the dataset by changing the key in corpus; valid keys = {'1'-'50'}

• Chord_Jazzification.py: either to train the models or to inference chord sequences using the pre-trained models

  The pre-trained models are saved in the directories: coloring_model, voicing_model
  The jazzifications of the JAAH dataset are saved in the directory: JAAH_inference

• chord_jazzification_models.py: the implementations of the models