I think I would want to find an existing sparse autoencoder implementation, ideally a project already setup for encoding audio, and start from there. http://www.stanford.edu/class/cs294a/sparseAutoencoder.pdf