부산대학교 도서관

Machine learning-based music generation has the potential to significantly change the music business and creative processes. The field of artificial intelligence offers new opportunities for creativity, artistic inquiry, and individualized musical experiences. Future music composition and generation will continue to be shaped by the research and development carried out in this area. Generative Adversarial Networks and reinforcement learning can be aptly used to improve the caliber and variety of music being created. The presented work discusses about the difficulties faced when creating music, whilst preserving musical coherence and preventing copying. A workflow is proposed for generating music using the LSTMGAN model. Loss graphs were used as the determinant of music quality. The results obtained show that LSTMGAN model performs remarkably well when used for music generation.