Abstract:
Recent years have witnessed great progress in using deep learning algorithms to learn to compose music in the form of a MIDI file. However, whether such algorithms apply equally well to compose guitar tabs, which are quite different from MIDIs, remain relatively unexplored. To address this, we build a model for composing fingerstyle guitar tabs with a neural sequence model architecture called the Transformer-XL. With this model, we investigate the following research questions. First, whether the neural net generates note sequences with meaningful fingering (i.e., string-fret combinations), which is important for tabs but not for MIDIs. Second, whether it generates compositions with coherent rhythmic grooving, which is crucial for fingerstyle guitar music. And, finally, how pleasant the composed music is in comparison to real, human-made compositions. Our work provides preliminary empirical evidence for the promise of deep learning for guitar tab composition, and suggests areas for future study.
