CompuSchmooze – October 2005: Dragon NaturallySpeaking is a well-spoken program
By Steven L. Lubetkin
Copyright © 2005 Steven L. Lubetkin. All rights reserved.
It’s been several years since we took a look at computerized dictation programs. I thought it might be a good time to revisit this technology, so I am writing this column without touching the keyboard very much.
I’m using a program called Dragon NaturallySpeaking Version 8, which is currently being distributed by ScanSoft (www.ScanSoft.com), better known as the manufacturers of a range of image scanning software programs and scanners such as the Strobe Pro single page scanner which we may get to in a future column.
Like its predecessors in this genre, Dragon NaturallySpeaking requires a fairly powerful processor. The manufacturer recommends at least a 500 MHz Pentium III processor, with 256 MB of RAM and 500 MB of hard drive space. The program only works on Windows XP or Windows 2000. There is a significant and noticeable improvement in its ability to recognize your words quickly and accurately since the last time I looked at this kind of software.
The last time I looked at voice recognition software, I had the amusing experience of having the words “bar mitzvah” rendered as “oat bran.” I’m happy to report that in dictating the last sentence, the program rendered the words accurately on the first try.
Matt Revis of ScanSoft said the company heard customer demands for greater accuracy.
“We decided that instead of doing a whole slew of bells and whistles around the new release, were just going to put almost all of our R&D investment into speech recognition accuracy,” he said. “We got close to a 30% improvement in speech recognition accuracy, and that’s just made a huge difference.”
Dragon creates a toolbar at the top of your screen with icons to turn the microphone on and off, and other drop-down menu picks for tools and instant training of unrecognized words. Before you actually begin working with the program, you do have to engage in a short reading exercise to train the program to recognize your voice correctly. However, the reading selections are short, and as far as I can tell, you only need to read one selection. Earlier programs like this required you to read multiple selections, taking as long as half an hour to train the program to be effective.
As you dictate, you vocalize the punctuation marks by saying their name. The program does recognize the names of the punctuation marks, but this can be a bit of a problem if you are tryingto dictate the word “period” or the word “comma.” To get the word to appear, I had to tell the program spell that, which opens a window where you can say the letters making up the word, by saying choose one or choose fiveand so on.
Dragon also works with a wide range of handheld dictation recorders, essentially any digital recorder that can create audio files convertible into the standard WAV file format. I haven’t had a chance to train the program to recognize dictation from my newest portable audio recorder, the Olympus DS2, but I will report on that effort in a footnote in a future column.
ScanSoft also distributes IBMs ViaVoice speech recognition program, but aims it at a less-sophisticated audience more focused on brand and more focused on price, said Revis. People who prefer Dragon are more focused on features, he said.
“The typical customer using Dragon is someone who is maybe more computer savvy, who wants to get more out of speech recognition,” said Revis. “The IBM buyer is more of an impulse buyer.”
There are customized versions of Dragon Naturally Speaking for medical, legal, and office professionals. Depending on which version of the product you purchase, the price ranges from $99 up to over $600 for the professional versions, but remember that these include specialized vocabulary/dictionaries for medical or legal terminology.
Be the first to comment