Audio to text only uses the first audio stream without any indication or possibility to change it.

[mzso]

I accidentally ran it on two files where the intended (english) track was the second one.
So I ended up with english subtitles generated from a non-english language track. (BTW it's surprisingly okay-ish but it, but I definitely didn't want to translate back to english in this case, where the original language is english)

I guess it's possible to work around it by extracting the track myself, but I don't see why couldn't SE could do this.