This repo contains code for compositional algorithms for autoencoder neural networks. My thesis proposes two algorithms, network modulation synthesis and predictive feedback, that are intended to more easily create interesting musical audio using autoencoder neural networks. This software here implements the algorithms using the CANNe synthesizer model as the generative model (the original CANNe codebase can be found here: https://github.com/JTColonel/canne_synth).
There are currently a lot of extraneous files in the repo. In the near future, I plan to clean up the code and also make the project importable as a Python library so other programmers can start creating music without having to clone this repo and install all of the dependencies!