TENT: Technique-Embedded Note Tracking for Real-World Guitar Solo Recordings
Journal
Transactions of the International Society for Music Information Retrieval
Journal Volume
2
Journal Issue
1
Date Issued
2019-01-01
Author(s)
Abstract
The employment of playing techniques such as string bend and vibrato in electric guitar performance makes it difficult to transcribe the note events using general note tracking methods. These methods analyze the contour of fundamental frequency computed from a given audio signal, but they do not consider the variation in the contour caused by the playing techniques. To address this issue, we present a model called technique-embedded note tracking (TENT) that uses the result of playing technique detection to inform note event estimation. We evaluate the proposed model on a dataset of 42 unaccompanied lead guitar phrases. Our experiments showed that TENT can nicely recognize complicated skills in monophonic guitar solos and improve the F-score of note event estimation by 14.7% compared to an existing method. For reproducibility, we share the Python source code of our implementation of TENT at the following GitHub repo: https://github.com/srviest/SoloLa.
Subjects
Automatic music transcription | expression style recognition | expressive music performance | guitar performance analysis | music signal processing | playing technique detection
SDGs
Type
journal article