Music Genre Classification Using Federated Learning

Abstract

Federated learning (FL) is a decentralized privacy-preserving machine learning technique that allows models to be trained using input from multiple clients without requiring each client to send all of their data to a central server. In audio, FL and other privacy-preserving approaches have received comparatively little attention. A federated approach is implemented to preserve the copyright claims in the music industry and for music corporations to ensure discretion while using their sensitive data for training purposes in large-scale collaborative machine learning projects. We use audio from the GTZAN dataset to study the use of FL for the music genre classification task in this paper using convolutional neural networks.