Jiaye Wang, jiw609@ucsd.edu
Fingerstyle guitar is the technique of playing the guitar by plucking the strings directly with the fingertips, fingernails, or picks attached to fingers, as opposed to flatpicking (plucking individual notes with a single plectrum, commonly called a "pick"). The term "fingerstyle" is something of a misnomer, since it is present in several different genres and styles of music—but mostly, because it involves a completely different technique, not just a "style" of playing, especially for the guitarist's picking/plucking hand.
For many musicians, guitar is like a small band. By hitting the guitar’s body, guitarists can mimic the sound of drum. By hitting the string in different angle, guitarists can also mimic the sound of different percussion instruments. Guitar fingerstyle develop this feature to a greater level. Hence, It will be interesting to figure out by combining machine learning and fingerstyle together, what kind of music are we going to produce.
This project will using guitar pro 5 software/ github code to convert guitar tabs in to midi file. Then the midi file will be sent into Music Transformer to generate new music sequence. After this, the two new formed midi file will be sent into Gansynth to replace guitar timbre to other timbre. I will also feed the original midi file into Gansynth as a reference.
In this project, I am going to Conny Berghäll - Revolution as an example.
Briefly describe the files that are included with your repository:
original guitar tab file: revolution.gpt5
orginal midi file: revolution.mid
Transformed midi file: 1-primer+continuation.mid 2-accompaniment.mid
Music Transformer: https://magenta.tensorflow.org/piano-transformer • colab: https://colab.research.google.com/notebooks/magenta/piano_transformer/piano_transformer.ipynb
GANSynth: https://magenta.tensorflow.org/gansynth • colab: https://colab.research.google.com/notebooks/magenta/gansynth/gansynth_demo.ipynb
Transformed midi file: 1-primer+continuation.mid 2-accompaniment.mid
Generated wav file:
1- generated_original.wav (produced by feeding original guitar midi file in to Gansynth)
2-generated_accompaniment.wav(produced by feeding transformed guitar midi file(machine generated accompaniment) in to Gansynth
In the Music Transformer, we are required to upload two midi file. The first midi file will contribute to the priming sequence of the resulted music, While the second midi file will contribute to the melody of the result music.
It will choose the highest pitch note as the meldoy. Although this is not accurate, we are still going to upload our original midi file into both section. Because we want to stick on our original guitar performence.
References to any papers, techniques, repositories you used:
Music Transformer: https://magenta.tensorflow.org/piano-transformer • colab: https://colab.research.google.com/notebooks/magenta/piano_transformer/piano_transformer.ipynb
GANSynth: https://magenta.tensorflow.org/gansynth • colab: https://colab.research.google.com/notebooks/magenta/gansynth/gansynth_demo.ipynb