Computers are nice. You can use them to play Tetris or send e-mail or switch from “Jason” to “Cassandra” in a chat room without anybody even noticing. Unfortunately, the damn things are still too much like the bulk of the species that created them: They just don’t know how to listen. Literally. That’s what led me to Dragon Systems NaturallySpeaking continuous-speech software ($695 Deluxe, or $160 for a smaller-vocabulary edition), the current front-runner in the voice-recognition race. The good news: I was able to speak into NS’s included headset microphone at rates of up to 120 wpm (my personal record) and watch the transformation of my voice into text. In fact, I transcribed this entire paragraph in less than a minute … and peeled a grapefruit at the same time. Now, that’s productivity!
NS didn’t even force me to talk like the 777-FILM guy to be understood. Instead, it asked that I select and read a 30-minute book excerpt (I chose Dogbert’s Top Secret Management Handbook) while the program analyzed my voice. After that, it used my daily writing activity to catalogue and “learn” from its initial misinterpretations of certain words (amazingly, NS rarely makes the same mistake twice). Out of the box, it hit the mark about 75 percent of the time, but after a month accuracy was up to a solid 96 percent.
The danger of all this, of course, is the threat of a verbal-diarrhea leakage into the written world and the further erosion of the distinguishing line between auteur and two random guys sitting around talking. In other words, good-bye carpal-tunnel syndrome and David Mamet.