Take a look at the great “Video speech matching” project at TheyWorkForYou.com. They’re combining BBC video of the British House of Commons and official text transcripts, and asking people to match keywords and phrases with the videos, to create a time-stamped archive of important debates.
Tom Steinberg, one of the organizers, says the community has responded in force, and people keep signing up to help. There’s no way to do this (at least not yet) with machines, he says, because machine speech recognition can’t (yet) handle all the different accents and dialects in one bite.
This is reminiscent of NASA’s wonderful ClickWorkers project, which relies on human perception to handle what machines can’t do as well. I love it.