rileynwong / lstm-jazz

generate jazz using lstm neural network

lstm-jazz

Generate jazz music using an LSTM neural network. See accompanying blog post.

Setup

$ pip install music21
$ pip install keras
$ pip install numpy 

Usage

1. Set training data: Create a folder containing midi files, or use one of the ones provided. ff_midi_songs contains music from Final Fantasy, herbie_midi_songs contains music by Herbie Hancock.
2. In train.py, line 20: for f in glob.glob('herbie_midi_songs/*.mid'):, edit the folder to the folder containing your training set of midi files.
3. Run $ python train.py
4. Once trained, you'll have a lot of resulting weight files, in the format weights-improvement-***.hdf5. The most recent one will be the one you want to use.
5. In generate.py, line 67: model.load_weights('jazz_weights.hdf5'), set the path to the weights file.
6. Run $ python generate.py. Your resulting song will be created in the same folder as output_song.mid

Credits

https://github.com/Skuldur/Classical-Piano-Composer