Abstract
Changing the genre of a song is one of the methods used when compositing music. To the best of our knowledge, musicians usually add their new ideas to the song, while trying to keep most of the special characteristics of the original song when arranging music. Similar to other artistic tasks that require human creativity, converting the genre of a song takes a significant amount of time and effort. In this project, we propose a method to translate a music genre by using machine-learning, which can generate a new song with comparably less amount of time than humans. Specifically, we utilized cycleGAN based model to translate a soundtrack to another soundtrack. Due to the complexity and difficulties of dealing with audio data, our model is able to handle files written with MIDI (Musical Instrument Digital Interface) specification only with specific characteristics. In the near future, we expect to expand our project to use regular audio files rather than MIDI to do our tasks to generalize our model. By doing so, we hope our model to be used for the general public without further modifications.