Thank you for the kind words, Sherlock! (And sorry that it’s taken me so long to be able to get anything exported!)
I think the whole mix is a bit like a Rubik’s cube. If one part gets changed, the other parts sound different by recontextualisation, even if they haven’t been changed themselves. It’s such a tricky delivery to work with as well, and there’s no good source for any of the exact phonemes we need.