I have been using Windows voice recognition with Windows 7 (given to me on a laptop by Yale, how nice) to copy research materials from paper copy into the computer. It's quite excellent but still erratic, maybe because the mike I'm using is an old VX1 that came with Dragon Naturally Speaking (version about 2) a long time ago. Still faster (for me) than typing -- since I can't look at the page I'm transcribing at the same time as I type (I'm a terrible typist) and between reading, switching focus to keyboard, forgetting what I read, switching back, remembering, typing, typing almost correct but in fact misremembered words, etc., etc., the VR is great.
But I have discovered a new use for the technology. I find that if you leave it on, with the microphone sitting on the desk, and do things like answer the phone, speak to someone at the door, root in a drawer, mutter imprecations under the breath, whistle-hum themes from Beethoven's 3rd, flush the toilet, climb stairs into the attic to look for something not there, the VR program will keep on listening, trying to make words of what it hears. It never types non-words, of course; all it knows are words. And I think the program has becomesophisticated enough to guess at grammatical strings -- it will guess a verb after a noun rather than another noun, etc.
Which means that it can form a sort of Delphic function, sorting through or distilling from your life actions, producing ambiguous messages for you to interpret. It has the chance function of I Ching, the mystic-utterance quality of prophecy, the secret-life-revealed quality of Tarot. Just leave it on all day (don't look!) and check in to see what you can learn, or seem to learn, or convince yourself you may or might have learned. This could be big: the transcription of non-verbal life into words without intermediary.