knut0815/neuralmelody

train and generate melody for pop music with recurrent neural networks

Overview

• This repository implements melody generation model proposed in this paper.

• The input is a two-hot vector in which the first 1 corresponds to a certain chord progression of 2-bar lengths (ex: C - Am), and the second 1 corresponds to the part annotation, e.g., verse, chorus, etc.

• The output is a MIDI file with generated melody converted from generated strings. Generated strings are currently in the form of pitch;pos;duration.

• This repository is a modification of NeuralTalk.

Dependencies

• pretty-midi

• hmmlearn

• mido

Usage

• To train

  python train.py

• To deactivate regularization on pitch range

  python train.py --reg_range_coeff 0

• To set pitch range for regularization (default is 60~72)

  python train.py --reg_range_min your_min_val --reg_range_max your_max_val

• To generate MIDI file

  python generate_midi.py cv/checkpoint_file

• To generate MIDI file with HMM-generated input (by default, song will be generated based on our pre-set test input)

  python generate_midi.py cv/checkpoint_file --gen_chords True

• Notes are inserted to MIDI files on a real-valued time instead of discrete musical lengths, so make sure to quantize it on any sequencer (e.g. GarageBand). 1/16 is recommended.

• Check our demos

Citation

@article{andrew2017neuralmelody, author={Andrew Shin, Leopold Crestel, Hiroharu Kato, Kuniaki Saito, Katsunori Ohnishi, Masataka Yamaguchi, Masahiro Nakawaki, Yoshitaka Ushiku, Tatsuya Harada}, title={Melody Generation for Pop Music via Word Representation of Musical Properties}, journal={arXiv preprint arXiv:1710.11549}, year={2017} }

License

BSD license.