Automatic Guitar String Detection by String-Inverse Frequency Estimation
T. Geib, Maximilian Schmitt, Björn Wolfgang Schuller · OPUS (Augsburg University) · 2017
In this work, we present a novel approach to approximating the fretboard position, i. e., the string and fret combination of guitar and bass recordings, using a feature we call String-Inverse Frequencies (SIFs). These frequencies are obtained from the opposite part of the string pressed down on a fretboard. We then show how they are calculated and proof their usefulness for guitar string detection. Additionally, a database is featured with recordings specifically tailored for this task. Furthermore, we demonstrate a basic approach using SIFs based on FFT spectral analysis and compare it to a basic standard classification process using Mel-Frequency Cepstral Coefficients and Support Vector Machines. The SIF-based approach showed a detection rate of up to F1 = 72% for both guitar and bass. Finally, we discuss further possibilities regarding SIFs.