Dictation software … that works

David Pogue of The New York Times has a weekly email column that is often quite useful. Today’s is especially so I thought.

In last week’s column, I described the weird and wonderful Comfort Keyboard, a three-section split keyboard whose universal joints permit total 3-D freedom of positioning. I mentioned in passing that I also use speech-recognition software to avoid typing and speed up my work.

As it turns out, more readers seemed interested in that part of the column than in my wacky keyboard. Many of you wrote for more details — and so here they are.

The program I use is called Dragon NaturallySpeaking 7 for Windows. (Mac fans always write to ask which program they should use. Truth is, the dictation software for the Macintosh isn’t anywhere near as good. I always suggest picking up some cheapo used PC to run NaturallySpeaking, and then transfer the resulting documents via network cable to the Mac.)

NatSpeak now comes from ScanSoft, which picked it up for a song when Lernout & Hauspie disintegrated in the European courts, thanks to embezzlement and fraud by its executives.

NatSpeak is available in a bunch of different versions, ranging from $60 to $200; you can see a feature comparison table at www.scansoft.com/naturallyspeaking/matrix. Each package includes a headset microphone, although aficionados who are really into accuracy replace it with a nicer U.S.B. model. (I use an Andrea headset with U.S.B. adapter from, for example, speechtechnology.com.) Ambient sound and coworker noise generally isn’t a problem, because the microphone is a half-inch from your mouth.

All NatSpeak versions offer the same accuracy, let you dictate into almost any program and let you both dictate and control your PC with voice commands (like “Close this window,” or whatever).

The Preferred Edition (which, in fact, I prefer) also lets you create voice shorthand. For example, when answering e-mail, I can say, “go away” to trigger a much longer response like this:

“Thank you very much for taking the time out of your busy schedule to let me know about how violently you despised my latest column. I guess we’ll have to agree to disagree this time around. After all, a healthy garden of opinion and independent thinking is what makes the world go round, n’est-ce pas?”

(Just kidding.)

You talk fluidly and normally—not—broken—up—like—this. You generally have to say the punctuation. (“Love, comma, new paragraph, David, period.”) The program nails homonyms like “to,” “two,” and “too” by looking at the context of your speech.

When the program does make a mistake (“the writer left” instead of “the right or left,” for example), you can correct it entirely by voice. You just say, “Correct ‘the writer left,’” and a little pop-up numbered list of alternate transcriptions appears just below the erroneous text. Usually, the version you wanted appears at the top of the list (it appears like this: “1—the right or left”). You just say, “choose 1.” Instantly, NatSpeak corrects the error, moves your insertion point back where you stopped, and teaches itself never to make that mistake again.

Now, if you could see me using NatSpeak live, you’d be floored. I routinely dictate a page or two without errors, at terrific rates of speed. My wife once clocked me at 120 words per minute.

But here’s the big “but.” Not many people actually use dictation software; the huge majority of people buy it, try it and never use it again. See, after “training” the software for the first time (reading a five-minute canned script), you get something like 95 percent accuracy. That’s one error out of 20 words, or several gaffes per paragraph. The program starts getting a lot better the more you use it. But you have to keep making those vocal corrections.

Within a couple of weeks, the software creeps closer and closer to 100 percent accuracy.

But most people, alas, simply don’t have the patience. There are so many times in life when an investment in time and learning up front leads to a long-term payoff. And in computing, that’s especially true (learning to use macros in Word, learning a few keyboard shortcuts in Mac OS X and so on). Dictation software falls squarely in that category.

For me, it’s a lifesaver and very nearly magical — but only because I stuck with it.

Visit David Pogue on the Web at DavidPogue.com.