I know that speech recognition probably is not quite good enough yet :-( I have another interest in this: since I am not a native english speaker, I find it often hard to understand everything that is being said in a video. English subtitles would often be helpful (especially for those videos of interviews or conference speeches etc.).

Perhaps Amazons mechanical turk could come in handy here? Would a clone of the mechanical turk specialised on movie transcripts be feasible? Or is it too special interest? (Perhaps most people either understand the words or they would prefer a translation?).

As for syncing subtitles: I don't know the details anymore, but I remember from ripping DVDs that the subtitles come simply in a text file along with the movie. So presumably something like you propose already exists.