Presented at ICASSP 2018.
Paper: https://arxiv.org/abs/1710.09064
Samples: http://srik.tk/speech-coding/
This code is very messy right now, and some work needs to be done to clean it up. More info about how to run it / reproduce results / etc. is coming soon. I believe all of the most important information (including hyperparameters) is in the paper, however.