I actually think there is room for a "human-assisted collaborative captioning" effort. You'd start with the results of automated speech recognition, and then you'd add in Google Docs-style collaborative editing of the closed captioning / fan subs.

Imagine if you could save a video by watching it and correcting the typos in the existing transcript.

But there's no way it'd be a viable business. (That's Rev, and it costs $1/minute, and it doesn't even give you a .srt file that you can upload to YouTube.)

Plus, you'd have to deal with spam/abuse, and creative uses of captions (e.g. niconico).

I tried building a prototype of this in college, but I couldn't figure out how to make it work financially, so I had to take a paid summer internship at YouTube.