We currently have this trained on 10 notes, we need to train it on much more data if we can expect a coherent sample of sheet music from this