304 — A Machine Learning Approach to Discover Rules for Expressive Performance Actions in Jazz Guitar Music

Giraldo & Ramirez (10.3389/fpsyg.2016.01965)

Read on 20 June 2018
#neural-network  #machine-learning  #jazz  #music  #musicality  #style  #guitar  #performance  #MusicXML  #sheet-music 

I just finished a guitar lesson where my teacher focused on tone and the expressiveness of certain patterns. There’s an art to it, to be sure. But like most of music, there are rules to it as well. I doubt machine learning could ever fully predict or identify expressiveness patterns in musical composition or performance, but it’s interesting to see if expressiveness can at least be detected.

This research identified a variety of modifications musicians made to the ‘written note’ in order to improve musicality, including alterations to tone, pitch, articulation, tempo, and other ornamentations. These are almost never notated in sheet music, but are still widely prevalent in jazz performance.

Conveniently, The Real Book (and other fake books) notate the “lead”, main melodic information, and core chord progression. Even though performers may not use this exact sheet music, it’s a good (and generally agreed-upon) foundation. This was converted to MusicXML notation, and aligned against the performance recording using an alignment technique that minimizes edit-distance.

This system outputs instructions for how to ornament music. The results include impressively intuitive suggestions: IF duration of note is very long THEN ornament note. Fair enough! Though there are some less intuitive-sounding rules: IF the tempo is medium (90–160 BPM) AND the note is played within a tonic chord AND the next note's duration is not short nor long THEN delay note.