Eksplorasi Deep Learning Menghasilkan Karya Musik Menggunakan Metode Generative Adversarial Networks (GANS) (Kasus Musik Genre Pop)

Abstract

Music artistry is an enduring form of artistic expression that continues to evolve across various genres. Among these genres, pop music stands out as particularly popular. Creating musical compositions is a challenging endeavor, requiring a profound understanding of musical notation, a skill possessed by select individuals, such as musicians. Even for musicians, a wealth of references is necessary to produce fresh compositions that can be appreciated by a wide audience. This study aims to explore the creation of new pop genre music using Generative Adversarial Networks (GANs). GANs, a widely adopted method, demonstrate the capability to generate novel works by leveraging two distinct components: the Generator and the Discriminator. These models engage in a competitive interplay, with the Generator striving to produce synthetic datasets that closely resemble authentic ones, while the Discriminator endeavors to discern between datasets generated by the Generator and genuine ones. Based on the conducted research, it is evident that GANs have the capacity to generate a diverse range of new music based on acoustic piano instrument notations, employing a dataset of 50 music files in .mid format.