Year: 2012
Type: conference-paper
Abstract: In this abstract, we propose a method to learn application-specific content-based metrics for music similarity using unsupervised feature learning and neighborhood components analysis. Multiple-timescale features extracted from music audio are embedded into a Euclidean metric space, so that the distance between songs reflects their similarity. We evaluated the method on the GTZAN and Magnatagatune datasets.
Authors: Sander Dieleman, Benjamin Schrauwen
Institution: Ghent University
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2294079921