I've been messing around with Higgs Audio that actually uses the delay pattern. It has to apply it and then unapply it after the generation. I noticed it's actually really hard to chunk and stream audio correctly when you need to apply and reapply these patterns essentially to the "entire" output.