mikus / musicgen

Music generator

MusicGen

Automatic music generator made during hackaton in Amazon Software Development Center Poland in cooperation with Jakub Sobotka.

Numerical approach

docker-compose run musicgen numeric generate -o <path_to_output_midi_file>

LSTM approach

Training

To start training:

docker-compose run musicgen lstm train start -n <unique_model_name> -p <dir_with_midi_files_to_learn>

To continue training (started earlier):

docker-compose run musicgen lstm train resume -n <model_name_used_previously>

Music generation

docker-compose run musicgen lstm generate -n <name_of_trained_model> -o <path_to_output_midi_file>