I have this inner voice when reading English, which is my second language. However, if I am reading stuffs in Chinese, I can skim through without the inner voice.
When I read Spanish (2nd language; not great at it) I sub-vocalize as I do in English. But if there is a long, complex word that I would have to laboriously "pronounce" (but I know exactly what it means ahead of time), I will treat it as a gestalt/symbol (so, it's like "cheating" on the vocalizing), so sub-vocalization isn't actually needed for comprehension, but I'm guessing is a parallel mental action/process (probably correlated with how we learn to read a language and personal characteristics). Maybe non-sub-vocalizing is really just suppressing this parallel processing workflow. (I'm wondering if people who "don't need to do this" could read backwords (accidental pun!) effectively.)