Testing Voice Recognition Software

(FSB Magazine) Brooklyn -- Many business owners regularly talk to inanimate objects. Don't believe me? I'm guessing that in the last week alone you've begged your PC not to lose valuable data or implored your notebook to recover lost documents. While we all have one-sided conversations with our tech toys, we generally don't expect them to answer, much less complete tasks simply because we say so. But today's voice-activated software promises to do just that, claiming faster speeds and an impressive 99% accuracy level.
Several years ago I fell and permanently injured the nerves in my right hand, so too much keyboard time can literally cramp my style. Hoping to ease the strain, I recently tested the latest versions of three popular voice-recognition software packages: Dragon NaturallySpeaking Professional 9 ($899), IBM ViaVoice Pro USB Edition, version 10 ($189.99), and the speech-recognition software included in the Ultimate version of Microsoft's new Vista operating system ($399). I tried all three systems on notebooks and desktops running various versions of Windows XP's small-business edition. I used them for everything from dictating e-mail to composing formal marketing pitches. Each system started easily at the click of an icon. But before they would follow my orders, I had to teach them my speech patterns by reading several scripts that appeared onscreen.

While I was generally impressed with their ease of use, I had to be vigilant about the results. Programs would sometimes slow down or speed up, resulting in repetitions as well as missed words. Punctuation was tricky; you need to say "period" at the end of a sentence. And I'm still wondering about the accuracy of those accuracy claims, because each system on occasion offered inventive substitutions for my actual words. ("Melody" became "nullity," and "man" transmogrified to "math," "Matt," and more.) Full disclosure: I admit to deriving a certain childish pleasure from testing - okay, tormenting - my would-be assistants by singing or using off-color language. Confronted with my favorite four-letter expletive, all three applications responded with prim substitutions such as "Oh, flock."

Dragon NaturallySpeaking Professional 9 (nuance.com/naturallyspeaking), which ships with an Andrea noise-canceling headset, is the latest incarnation of the Nuance desktop speech-recognition software formerly known as ScanSoft. The new version features an appealing interface (who doesn't love dragon icons?) and includes a cheat sheet with the most common commands. Dragon allowed me to choose between two narrator voices (I nicknamed them Computer Lady and Earnest Guy), which provided clear instructions. I was told to speak slowly and carefully, like a TV anchor reading the nightly news.

IBM's ViaVoice application (nuance.com/viavoice) came with a simple setup disc, plus British and American English options. ViaVoice ships with the same headset as Dragon, along with a USB connector. Despite initial compatibility issues, I set up the software in less than an hour, and the cheat sheet helped with basic commands. But ViaVoice was slower to establish voice patterns than either of its competitors. It also featured the most irritating icon of all the applications that I tested. Woodrow, the talking pencil? I mean, really. After ten seconds of Woodrow's virtual company, I was tempted to snap the little flock.

Vista's speech-recognition software was the easiest application to set up, although it provided no tips or tricks. Vista picked up most of my slang and intentional mispronunciations. But it seemed to work better with an external headset microphone than with the internal one in my Vaio notebook. Worse, I have so far been unable to banish the software from my desktop If forced to choose a favorite, I'd probably pick ViaVoice (Woodrow notwithstanding) for its excellent dummy-proofing. Still, I think any of the applications could be a gift to entrepreneurs with repetitive-stress disorders, arthritis, or impaired vision, as well as users who are strapped for time or who type more slowly than they speak.

I got an interesting article from CNN.com to share with you guys. I personally think that voice recognition system will be the next issue for the IT development. Not only recognizing simple talks, which we can experience in car navigation and cell phone's voice recognization program, but also in near future, we might be able to use the internet without using the keyboard. Interesting huh? Share some ideas with me.