Exploring the Classification of Pipe Organ Sounds: Recognizing Instruments, Organ Types, and Registrations

Abstract

Classification of pipe organ sounds is challenging as the pipe organ has a huge variety of timbres depending on the selection of stops and the type of the instrument. Recent advances in foundational models for audio enable the exploration of classification tasks for pipe organ audio recordings. We focus on Organ Type Recognition, Organ Identification, and Registration Recognition (both intra-track and for full tracks). Employing BYOL-A embeddings and the multi-layer perceptron, we achieve impressive classification results.