Looks interesting. Haven’t watched all videos yet, only the introduction, overview and voices ones.
Did I understand this right?
a) needs processing time, thus introduces a delay, so not for realtime play
b) it tries to really identify melody lines in polyphonic music and assign each musical voice to a midi voice - but may be wrong and then corrects assignment as soon as the mistake is spotted retrospectively. This only works if the delay is big eneough that it can correct stuff so far in the past.
Sounds highly interesting! Not sure where I would use it though, given the (understandable) latency constraints.
Might still be worth playing around with a little! Not cheap, also, so should be better be used extensively then 