My God. It's like I think, "It would be interesting to work out X ..." and a few months later DeepMind's announcing some results in the area. Cool / Frustrating.
And better results than you could possibly expect. I was just getting into playing with audio synthesis, but I figured you'd always need to use spectrograms and get distorted data--then they dropped WaveNet, because screw that, we can just process the raw waveform!