A Comparison of 3 Deep Learning Music Tools

Deep learning is becoming a more and more popular subject matter but what are some the ways in which creatives have explored these new trends? Today we are going to mention a few of these more creatively focused applications and will include links to GitHub as well.


The team over at Google has been busy on this one. The project became open source in 2016. They are still working on improving the models. Models are generated by submitting thousands of midi files. After training Magenta with new models new pieces can be instantly generated.

GitHub: https://github.com/tensorflow/magenta

Below is an example.


This is a project by Feyman Liang at Cambridge University and its goal is to generate music modeled after a great musicians work. Ultimately, the projects seeks to emulate Bach so well that its compositions will be indistinguishable from the real thing.

GitHub: https://github.com/feynmanliang/bachbot

Below is an example.

Flow Machines

A research team in Paris is responsible for the development of this system. Unlike the systems previously mentioned, this one is more or less intended as a way to get the creative juices flowing. It seems less intent on replacing a musician and more focused on aiding one. The system they have developed can generate any sort of song from its database. Currently, with over 13,000 examples for reference it also represents the first time an AI wrote a pop song. Even if it still needs humans to play and record all those instruments. Below it a link to a “Beatles” style pop song.

GitHub: Sorry this one isn’t open source.

Below is an example