I usually set subtitles if there is such a strong accent in the acting that I can't understand the words. But then, I wouldn't know if the subtitles are complete or not, would I?

I've subtitled some academic videos at the university library; getting the exact words on the screen at the right time is a lot of work. My boss took an American video, just a couple of short interview clips, to a Chinese conference she was presenting at. It was maybe eight minutes total, and it took three of us a couple of months. The first person hunted for the best and clearest version of the videos and made the clips, then I transcribed them. I sent these to our Chinese-speaking librarian to translate into Simple Chinese. I then had to drop those symbols back onto the video and return them to him to check. In some cases I had too many or two few words on the screen at a time, but didn't know it because I don't read that language. We went back and forth a couple of times to get that part right (and then the dean never told us how it went and never said thank you for all of our troubles!)

I also transcribed some of the sports videos for campus basketball games played by the wheelchair athletes for their archive. Boring watching the game and listening for calls, and in many instances going back over to figure out what they said (we're talking announcers over loudspeakers on the video) before committing it to text on the screen.

There are automated transcriptions and those are notorious for grammatical errors, because of the choice of the wrong homonyms for common words. Machines don't do syntax very well.