That’s where the onset descriptors comes in
I did some recordings to test stuff further too, to see if I can get to the bottom of the loudness discrepancies, and to also test out the pitch descriptor effectiveness, though I long for some descriptors-land sigmund-ing.