Easy-to-use Deep LSTM Neural Network to generate song like containing improvisation.

Demo (SoundCloud)

• Keras
• TensorFlow
• Python MIDI

pip install -r requirements.txt

python ./src/parse_midi_to_text.py

python ./src/training.py

python ./src/generate_music.py

• You can use other MIDI file to train the model to generate new song. Change the file ./midi/original/original_song.mid.
• MIDI format is usually consist of multiple track and this repository is currently not supporting automatic detection which track is main part of the song. So you may have to choose track as index of the pattern in parse_midi_to_text.py.

MIT © Tatsuya Hatanaka