Automatic String Detection for Bass Guitar and Electric Guitar

Abstract

AbstractIn this paper, we present a machine learning-based approach to automatically estimate the fretboard position (string number and fret number) from recordings of the bass guitar and the electric guitar. We perform different experiments to evaluate the classification performance on isolated note recordings. First, we analyze how the separation of training and test data in terms of instrument, playing-style, and pick-up setting affects the algorithm’s performance. Second, we investigate how the performance can be improved by rejecting implausible classification results and by aggregating the classification results over multiple time frames. The algorithm showed highest string classification f-measure values of F = .93 for the bass guitar (4 classes) and F = .90 for the electric guitar (6 classes). A listening test with 9 participants with classification scores of F = .26 and F = .16 for bass guitar and electric guitar confirmed that the given tasks are very challenging to human listeners. Finally, we discuss further research directions with special focus on the application of automatic string detection in music education and software.Keywordsstring classificationfretboard positionfingeringbass guitarelectric guitarinharmonicity coefficient