lgtm - some thoughts on an adjacent issue.

This was not detected by yaml tests, although they consider data preparation scripts - why?

While the recipe overview has the data preparation script referenced, it plays a litter role for the yaml testin.
https://github.com/speechbrain/speechbrain/blob/develop/tests/recipes.csv

Should testing include that through YAML one could specify parameters of any subfunction?
(e.g., retraversal of uses; namespaces, and if all function arguments are settable then; their default is changeable through YAML.)

This was not detected by yaml tests, although they consider data preparation scripts - why?

Hmm according to the doc string of

Not sure to understand clearly your last paragraph.

this is "normal" that the tests have failed to catch this problem because the tests are only checking if a variable is unused in the python files while in our case we just forgot to declare it and use it. (which is a different problem)

You got it; don't worry. I'd see that as a part of functionality for that testing script, whereas it is currently not. It's literally up to us to make it function/or not. Yet the existence of this PR demonstrates: it should be part of function. Question is: what's the least effort here - writing a new testing script; reusing the old one; some hybrid approach - or simply fixing this instance and waiting until it pops up elsewhere again 😉