Do the APIs support simultaneous voice transcription in a way that different voices are tagged? (either in text or as metadata)
If so: could you split the audiofile and process the latter half by pitch shifting, say an octave, and then merging them together to get shorter audiofile — then transcribe and join them back to a linear form, tagging removed. (You could insert some prerecorded voice to know at which point the second voice starts.). If pitch change is not enough, maybe manipulate it further by formants.
Yup, the influences on e.g. Newton happening to delve into reading up on Archimedes, Descartes, Fermat, and then synthesizing their inventions in his mind with lot of time on his hand, or for that matter Leibniz getting math tutoring from Christiaan Huygens seem to be crucial in relation to the invention of fluxions/infinitesimals. (Approximately from memory of reading Infinite Powers by Steven Strogatz).
I kinda like what you are saying, but it's not that clear-cut in my experience / skill-level.
One can write code that would work correctly, but trying to express the types as precisely as one would like may be painful and doesn't necessarily be productive.
Sometimes when I feel it is cumbersome to express my intent in typing,
I may feel satisfied enough to make a runtime check, (and perhaps add a note but depends...).
Especially if that functionality is used in scoped area (e.g. module) and not exposed to the rest of the app.
When trying to make a generic and flexible functionality, the typing experience seems get harder and more time-consuming for human to grok, and it may not even be that useful when you look how the code is used.
So I usually try to make typing as good as I can by finding documentation related to what I'm trying to do and banging my head to the wall from a couple of angles, and for some specific parts I may give up and type any.
Perhaps sometimes when typing feels hard it may tell that the abstraction is just plain wrong, so there's that plus side!