Speaker: Alejandro André Vivas Freitas.
Abstract: Music can evoke countless emotions regardless of culture or age, and although the classification of musical genres has existed for centuries, automatic classification is a relatively recent discipline (barely two and a half decades old) that has made significant progress thanks to machine learning and deep learning techniques. This study focuses on the specific task of classifying metal subgenres, analysing their history and sonic characteristics. By evaluating various machine learning models, ranging from support vector machines (SVM) to deep learning architectures, the system achieved 65% accuracy on a metal dataset and results comparable to the state of the art on the GTZAN dataset. Despite the high sonic similarity between subgenres, their frequent fusions and the subjectivity involved in labelling, this accuracy rate is considered satisfactory and represents a substantial improvement over previous studies.