This is fantastic. Selection of sounds is very interesting. It sets your brain on the quest of transderivational search, traversing his own spaces to find (derive) meaning when hearing these snippets of speech. They are ambiguous enough to really f$Y^k it up a bit, so process unfolds a bit more than usually.
Maybe currently looped squares in one column could not overlap, but that would be needed if this was some mixing table and it is not (it is a bit). It's awesome!
I really would love to hear some back story on why (eg. sounds selection).