Eavesdropping on Computer Speech

Related link: http://songcarver.com/C151287817/E1050038590/index.html

I got a tip about Plasq’s comic-book creation software Comic Life today, and noticed that the company founders were all musicians. So I started exploring their personal sites. On one, programmer Keith Lang poses the intriguing question, “What if computers talked out loud to each other? What if we could learn that language?”

He talks about listening to his modem and wondering what it would be like to address computers idiomatically, the way a composer would write differently for a French horn and a guitar even though the pitch ranges overlap. He even includes sound files and frequency plots of potential conversations.

I’ve read that telephone voice-recognition systems decrease in accuracy as callers try to speak more precisely, because the computers are trained to decipher sloppy speech. Ironically, speaking more precisely is the natural reaction when someone doesn’t understand you.

Seems to me that we already go out of our way to talk to computers on their own terms. Perhaps the answer is to make them more active listeners, asking more explicitly for the information they need and working with a range of answers instead of rejecting all but the “right” one. Ray Kurzweil told me [Amazon “search inside this book” link] that the hardest thing for a musician is to be an accompanist. “To be an intelligent accompanist means really understanding what the singer or violinist is doing, and then trying to strengthen them if they’re reaching a weak point,” he explained.

The message is clear: Not only must computers and humans speak more clearly, both need to listen harder as well.

How do you get your computer to listen to you?