One of the subfields of MIR (Music Information Retrieval), music genre classification, is becoming more and more popular among scholars, largely because there are still many unresolved issues. Despite the fact that there have been many published works on the subject, the base of the research still has a flaw: there is no formal definition of what a genre is. Music classifications suffer from human subjectivity and a lack of consensus, making them hazy and confusing. This essay's first section is a survey meant to address the topic's several facets. Its main objective is to provide the reader with an overview of the background and the present state of the art, while also examining the methods and datasets that have been employed up to this point and highlighting current issues, such this uncertainty, such as the adoption of human-centric strategies or the uncertainty of genre classifications. The paper gives particular emphasis to recent developments in machine learning as they relate to the issue of music annotation. Finally, using the Audio set, we examine various machine learning models in a music genre classification experiment.

Classification, MIR, Audio Feature Extraction, Machine Learning, DCT.