Voice Recognition Software Put To Test
Voice recognition software has been around for years but until a few weeks ago, I didn't have a strong personal interest in the topic. After breaking my elbow I suddenly found myself physically challenged, unable to type with both hands. After some healing and a new cast, I can type again but during the interim I had both an incentive and an opportunity try out Dragon Systems NaturallySpeaking 9 as well as the speech recognition feature that will be included in Vista, the next version of the Windows operating system that's expected to be available early next year.
For the most part I found both the Vista application and NaturallySpeaking to be accurate and easy to use though now that I can type again, I prefer using the keyboard.
To their credit both programs are accurate and intelligent in that they recognize not only words but also allow you to issue commands. And because the computer is doing the typing it's less likely to misspell words than I am. There's no question that the Preferred ($199) version of NaturallySpeaking is more sophisticated than the free voice recognition application that will ship with Windows Vista. It has support for multiple languages and special vocabularies and works with portable dictation equipment. Nuance's $99 standard edition also does a good job recognizing speech and lets you launch, switch between and control programs with the sound of your voice. Both versions come with a headset that does a good job picking up your voice.
There is also a professional edition for corporations and agencies that allows users to distill multiple commands into a single word or phrase. There are also optional dictionaries for lawyers, doctors and others who need to deal with specialized vocabularies.
Both the Microsoft and Nuance products allow you to get started without having to spend an extensive amount of training the software to recognize your voice. I found that NaturallySpeaking was pretty accurate right out of the box but the more you use it and correct its mistakes the better it becomes. The same is true with the Microsoft product. After using both for a few days, each got to the point where they recognized about 99% of my words as long as I spoke clearly into the microphone or headset. Both allow you to correct mistakes by speaking "select" followed by the erroneous word and then "correct that." You then get a menu of optional correct words which you can select and each time you do so, the software remembers the correction for next time.
At the moment, I'm writing—or should I say "speaking"—this part of the review using the application in Vista. When I first started using this program it was hard to get through a paragraph without at least one mistake. And now I'm sort of getting about 99% accuracy, which actually isn't that bad. I say sort of because while the program is making very few mistakes I continue to make them so I do wind up having to make a lot of corrections because I can't in this speak more often than I missed type. One might not notice that in conversation or even on a radio or TV interview but you certainly would notice it when you're reading.
Janet Kornblum, a technology writer for USA Today has been using successive versions of NaturallySpeaking for several years. Although she sometimes does use the keyboard, she prefers to "speak" her articles and email because she suffers from a repetitive strain injury that makes it harder to type. She is very bullish on the software but admits, "it forces me to read my writing because it often contains embarrassing 'speakos' as I call them -- missing words or plurals when I meant to be singular." She said that the software "has improved dramatically over the years," but , she tells some funny stories about mistakes which, if not corrected, could have gotten her into trouble.
In addition to training the software, Kornblum says, "I have had to train myself how to write by speaking which is really different than writing by typing." I've noticed that as well. There are different thought processes that go into speaking and typing.
There's also the issue of fatigue. While talking is arguably easier than typing, I personally find myself getting tired more easily when I speak than when I type. Perhaps that's because I've been typing for many years and am new at talking to my computer. However, with my sudden temporary disability, the equation has shifted towards voice. And of course that's the point. Not everyone has access to both of their hands and not everyone is an accomplished typist.
Having been temporarily disabled, I can easily see how this could be a wonderful tool for anyone with a permanent or long-term disability as well as those suffering from repetitive strain injuries or other hand problems. Did you notice the hyphen between "long" and "term." The software was smart enough to put it their(CQ). Oops. I guess it wasn't that smart. It just typed "their" instead of "there." And there lies the problem. Computers are not as good as humans when it comes to context. For example, if I were talking about the Internet and used the word site you would know that I'm talking about the web site – but the computer might not know whether I meant "site" or "sight" or "cite." Still, it does learn from its mistakes, which is more than I can say for some humans.
By Larry Magid