It's a shame the article uses soundcloud clips, because I suspect the end sounds aren't the impressive bit. I think it's the user interface that's the interesting bit here.

> In addition to the NSynth “slider” that Engel recently demonstrated at Google headquarters, the team has also built a two-dimensional interface that lets you explore the audible space between four different instruments at once. And the team is intent on taking the idea further still, exploring the boundaries of artistic creation. A second neural network, for instance, could learn new ways of mimicking and combining the sounds from all those instruments. AI could work in tandem with AI.

A video of this would have been much more useful to me.