It would be really nice. The problem is to generate the transcript. Even YouTube speech recognition system is crap when there's not a perfect, clear sound. Maybe you could use the transcripts that are already on YouTube when the uploader has created it, but of course it wouldn't be so useful.