NON Beer related - How Does Voice Recognition Technology Work?

***** Warning: NON BEER RELATED *****

This was a Nuance Press release today that I wanted to share with everyone...

How Does Voice Recognition Technology Work?

You might not realize it when you’re on the phone waiting for a computer to figure out if you just said “Flight Three” or “Fight me,” but it’s kind of a miracle that speech technology works at all. Wanting to know how it works, we called Dan Faulkner, director of product management and offer marketing at Burlington, Mass.–based Nuance, which has programmed speech recognition software for Amtrak and Sony’s PlayStation.

First, software has to figure out when you’re talking, and when you’re done. “Speech has certain characteristics, certain harmonic frequencies, and the software waits to identify these,” says Faulkner. When a computer identifies speech and then hears a pause, it sends that snippet along to the next stage of the program, which tries to decode it.

“The next stage is all based on phonemes,” says Faulkner. “Phonemes are the basic building blocks of speech, the sounds we make to make the words. So a word like ‘thought’ is broken down into three phonemes, which you might spell as th, aw, and teh.”

The computer looks for the phonemes and then, in trying to decide what word you’ve said, compares it to a database of words broken down into phonemes. The comparison, though, isn’t just a straight search, but a probabalistic statistical model, based on the computer’s idea of what you’re likely to say next. “If you’ve just said the, for instance,” says Faulkner, “the computer isn’t going to look for a verb, because a verb very rarely follows the.”

And if all goes correctly, the voice recognition program will have a fair chance of actually figuring out that you really do want to fly to Mexico City instead of New Mexico. — B.U.

link to article