Musical-style-transfer-with-cycle-consistent-generative-adversarial-neural-networks

A Master thesis by Simeon Ivanov Kraev for Osnabrück University - 01.2019

Atempting artistic style transfer basen on https://arxiv.org/abs/1703.10593 CycleGAN model by using spectrograms and Mel spectrograms as the networks input.

This work explores whether musical style transfer is achievable using the CycleGAN architecture with spectrograms instead of pictures. Whether the spectrograms make a good input format and does the length of spectrograms matters for the mapping between sets.