v0.1

This model is trained via extra conditioning on its own input. The next version of the model (v0.2) will have a slightly different architecture that seem to be providing deeper restoration, especially of narrow band audio.