Variational Convolutional Autoencoder originally developed for parsing the MNIST dataset adapted to parse feature matrices characterising sound signals.
https://debuggercafe.com/convolutional-variational-autoencoder-in-pytorch-on-mnist-dataset/
Note: add your own directory paths to the respective lines in train.py; This NN has been adapted for NPY data with 24x24x2 tensors, it has a custom data loader and a custom 'clear all (variables)' function run automatically before training begins.