An Unsupervised Methodology for Musical Style Translation

Deep learning techniques applied in image style transfer have shown great success. In the domain of music, this task can further aid musicians and music-generating AIs in the production of music. In this paper we present an unsupervised methodology for musical style transfer, which do not rely on strictly paired data to train. The model is capable of translating the style of symbolic music from the source domain to the target domain while mostly preserving the content and structure of input data. Output samples are evaluated by genre classifiers and show promising results.