Github Link: https://github.com/allisR/CS5647_Text2PianoMusic.git
- Generate dataset. Run
preprocess_midi.py
in theutils
folder. This will write pre-processed data into thedata
folder split intotrain
,val
, andtest
as per Maestro's recommendation. - Run
train_decoder.py
to train the audio decoder. We provided the pre-trained weight file asall_best_acc.pickle
. - Test the trained audio decoder by running
decoder_generate.py
. The output midis should be in theoutput_midi
folder asprimer.mid
andprimer_after.mid
.
- Run
train_decoder.py
to train the whole model. We provided the model weight file asall_best_acc.pickle
. - Run
model_generate.py
to test the whole model and generate a midi file based on the text input. The output midi should be in theoutput_midi
folder astext.mid