Matthew Gibson Undergraduate Dissertation 2014/15

Learning Guitar Fingering from Tablature

Supervised by G.Brown

Abstract

The intention of this project is to write a piece of software that will learn, through the use of neural networks, the correct fingering for a piece of music from guitar tablature. The project will scrape a large amount of tablature from the internet, pre-process this into a standardised format and then convert it into note and octave values. This data will then be passed into a neural network where the software will learn the correct fingering for each piece of tablature. For the outcome to be accurate, the neural network will have to have context, be trained with a large amount of data and then tuned using more data. As notes on the guitar occur in many places and across five octaves along the fret board, the challenge is to keep the fingering within a certain range of frets to ensure playability. It should be possible for the resulting program to produce fingerings for any sequence of notes it is given.