...for someone to at least tell me the kind of device I'm about to describe is possible. Within the rest of my lifetime. For less than a small fortune:
I'm a reporter. I skipped third grade, so I'm sure I missed some handwriting classes. Ever since, my handwriting has sucked, especially when someone is speaking fast and I'm trying to get down every interesting word. It's sort of a self-encryption - if I don't transcribe the notes within 12-24 hours, even I can't make out what it says. (It's why I love phone interviews - not out of sheer laziness - well, not just that - but I'm a real fast typist, so those notes are 1,000 times better.)
Sooo.... there's voice-recognition software out there, right? Getting better all the time? But apparently you still have to train the thing? I'd LOVE to tape interviews, rather than scribble them, then upload the audio file and have a program turn it into text! Heck, I'd settle for 30 percent accuracy, that'd be better than my stinkin' notes!
My friend/co-worker Jesse told me it'd take more processing power than anyone has at the moment. But "Utterly Boring" Jake agrees with me that'd it be so useful to so many folks.
So ... am I dreaming? Will it arrive in another 50 years, too late for me? Or ...
Oh and another invention idea, my wife Deb and I had - why not sell scrapbooking software IN a scrapbook? (And photo album software IN/with a photo album, for that matter?) With the pages and the software set for those pages? Clip art built in, the whole package together, "hardware" and software? I've thumbed through scrapbooking magazines and haven't seen that, have I missed it? (I should patent that one, yeah right;-)
3 comments:
I say we just cut to the chase and invent an inventing machine, and then let it complain to itself with adages like "If I can send a man to the moon why can't I make a decent Briscuit Sandwhich?" and the like. ;)
I'm sorry, but I did have to insert this little aside as well.
At Bend.com in 2001 we were playing with some Microsoft speech recognition software designed for gaming. Yes, you had to train it. Yes, it had a vocabulary of like 2 dozen entirely pre-built phrases. You say something, it repeats what you said in a Majel Barrett sort of way, and then carries out the dead-simple order.
We go through the little training wizard and try playing Starcraft, the first thing we try to get it to do is to select infantry group number seven so that we can move them out. We say "Go seven" into the microphone. after an odd 3-second long pause, do you want to know what it said back to us? "Build Barracks". Then all of our little villager units swarmed out to begin construction on some ill-conceived project behind enemy lines; where we were trying to send group seven.
"Go Seven" and "Build Barracks" are not 30% similar. Not at all. Not when Majel prounces it "bear - racks" and the vowel points don't even match up. This was in a quiet room, a deliberately pronounced command specificly to be easy on the machine, point blank into a microphone.
So, take your tape recorded out into street noise, with sirens and PA systems, or into a crowded courtroom and have some fellow lay a thick british accent on your mike, interspersed with "um"'s and "uh"'s, and let ol' Microsoft Speech-to-text get ahold of it and you'll likely wind up with MacBeth or The Pickwick Papers.
My grand idea for the evening is a speech-to-text interpreter that would have at least eighteen times the accuracy of these sorts of systems, without even having to be on location. I call it "A thousand monkeys on a thousand typewriters". :)
It's not the awesome invention you're asking for, Barney, but David Pogue, tech writer for the NYT, raves in a recent column about this speech-recognition software NAT8. Yeah, you have to train it, but you got 5 minutes to do that, right?
Post a Comment