Reviving Magenta’s Tensorflow Library

Recently I’ve completed my thesis outlining my work updating an outdated tensorflow library used for many early neural audio synthesis projects.

This project served as my final thesis for my Master’s degree at Maynooth University. My initial goal for the project was to implementing some of these earlier projects into a distilled plugin to highlight the limits of neural audio in real-time plugins. My target repository for this modification was Google’s Magenta project, a large repository with many intersting neural audio applications. However, the Magenta repository was based on Tensorflow 1, which has since been deprecated, leaving future audio tinkerers bereft. Thus, my goal for this project evolved to reviving the Magenta framework to make use of the modern Tensorflow 2 paradigm. A full report of the project, as well as the final results, will soon be available under my projects page.