Song2Guitar: A Difficulty-Aware Arrangement System for Generating Guitar Solo Covers from Polyphonic Audio of Popular Music

This paper describes Song2Guitar which automatically generates difficulty-aware guitar solo cover of popular music from its acoustic signals. Previous research has utilized hidden Markov models (HMMs) to generate playable guitar piece from music scores. Our Song2Guitar extends the framework by leveraging MIR technologies so that it can handle beats, chords and melodies extracted from polyphonic audio. Furthermore, since it is important to generate a guitar piece to meet the skill of a player, Song2Guitar generates guitar solo covers in consideration of playing difficulty. We conducted a data-driven investigation to find what factor makes a guitar piece difficult to play, and restricted Song2Guitar to use certain hand forms adaptively so that the player can play the piece without experiencing too much difficulty. The user interface of Song2Guitar is also implemented and is used to conduct user tests. The results indicated that Song2Guitar succeeded in generating guitar solo covers from polyphonic audio with various playing difficulties.