If there is no sample rate conversion then I would look at what the above suggests and make sure that time stretch isn’t happening on a global and per machine basis.

I’d be interested to hear your thoughts as is I get mixed results with some samples, particularly weak kicks.