DARPA is the government’s weird science lab, and they’ve got all sorts of random projects cooking. Among them is an attempt to improve speech recognition with the help of, well, pretty much everybody.
DARPA’s idea is to crowdsource speech recognition. As anybody who’s used a Kinect or Siri knows, speech recognition has substantially improved from even a few years ago, but it’s still pretty awful; you can’t use slang, for example, and if you’re even slightly muffled, the system won’t pick you up.
The basic idea behind the technology is that in addition to modified speech recognition programs installed in cell phones, humans would also be listening in and transcribing. Not just any human, of course, but a network of people certified to do this.
It actually makes sense, because automated speech recognition actually needs a human to carefully scan and correct it, so you might as well have the human listening in. The result would be a fully searchable text of, if DARPA gets its way, everything you’ve ever said into a microphone of any sort.
…OK, maybe this is a bad idea.