Deep neural networks: A case study for music genre classification

Abstract

Music classification is a challenging problem withmany applications in today's large-scale datasets with Gigabytesof music files and associated metadata and online streamingservices. Recent success with deep neural network architectureson large-scale datasets has inspired numerous studies in themachine learning community for various pattern recognitionand classification tasks such as automatic speech recognition, natural language processing, audio classification and computervision. In this paper, we explore a two-layer neural network withmanifold learning techniques for music genre classification. Wecompare the classification accuracy rate of deep neural networkswith a set of well-known learning models including supportvector machines (SVM and '1-SVM), logistic regression and '1-regression in combination with hand-crafted audio features fora genre classification task on a public dataset. Our experimentalresults show that neural networks are comparable with classiclearning models when the data is represented in a rich featurespace.