This project is an attempt to build a simple MIDI generator using RNN. The project is written on Pytorch with FastAI. For evaluation purposes only Pytorch is needed if IntToFloatTensor will be implemented. I used IntToFloatTensor from FastAI to have an ability to easily show rolls in notebooks.
I use several different metrics to evaluate a quality of my models. You can see their values in my notebooks.
To gerate new MIDI files you need to pass as arguments gan.py a beggining part of your MIDI from data, a path where you want to save output and a time length which you want to append to the beggining. Like
'python3 gan.py data/name.pmg midi/name.midi 100'