[ebj73]

A nice idea, but it sounds as if they're not quite able to get the input latency low enough. It needs to be around 3-4ms or less. Here it seems as if it's more in the 10-20ms range, judging from the demonstration videos. It's enough to throw a performance off, and to make everything sound just slightly behind.

It could be that it's not the input latency itself, but rather the tapered nature of the wind instrument attack curve. If a tapered attack curve is used to set of the playback of a sample, that could in itself also introduce extra input latency. It's not entirely clear what's happening here.