Can machines generate music? Recent deep learning-driven systems have made breakthroughs on this topic. Some famous among them are Coconet, DeepBach, GLSR-VAE, C-RNN-GAN, etc. However the generated music is still not satisfying based on the feedbacks from different music communities.


 


In this talk, we will analyze and compare various systems and experiments proposed by various researchers, illustrate their highlights and limitations through several examples.


Then we will introduce our new composer and discuss some possible solutions to several main challenges in this area.