Close, but not there
The day when users will be able to speak to computers and have their words appear on screen is getting nearer, but don't count on it yet.
By DAVE GUSSOW
Published August 29, 2006
Dragon NaturallySpeaking Professional 9
Company: Nuance System: Windows
Price: about $700
Speech recognition software is a lot like dealing with your kids: Sometimes it listens, sometimes it doesn't, and sometimes it has a mind of its own.
It also takes patience; yelling doesn't help.
Speech recognition and the idea of talking and controlling computers by voice has been a Holy Grail for the technology industry. Yet it still has a long way to go before people will have the same experience they've seen in Star Trek for almost 40 years.
A couple of recent developments gave me an opportunity to once again try speech recognition. Dragon NaturallySpeaking version 9 was released, and an injury made using a keyboard difficult. While still not perfect, the test was far better than previous efforts about four years ago.
Another thing to note: Many of the programs I used years ago no longer are on the market, an indication of just how difficult mastering this technology is. It's also not cheap. The professional version I tested has a list price of about $700. I found consumer versions, likely with far fewer features, online for $90 and $165.
I talked to a couple of people who have been using such software regularly and over a longer time than my periodic tests. Jim Roach, 54, of Cape Coral has been using it since the mid 1990s when GM was doing research to use speech recognition to help mechanics navigate electronic manuals. Stan Segien, 52, of Largo has been using it for about six years.
"They shouldn't call it talking to your computer," Roach said. "There's a big difference between how we talk and how we dictate."
Roach says he's a good typist at 80 words a minute. Not so for Segien.
"I'm lazy," Segien said. "If I had to sit there and type e-mails, I probably wouldn't do it."
After a couple of weeks of practice, I was ready to dictate this column. The goal, which I didn't achieve, was to write the column and send it to the office without touching the keyboard or mouse. Some corrections became too frustrating with voice. Getting the e-mail and Web addresses right was a challenge, so I caved.
In the beginning, installation and setup took about 45 minutes. It required some reading, mostly for microphone checks. The microphone has to be placed just so, but the process was much easier and faster than it was four years ago.
Roach and Segien disagree on the microphone. Roach thinks the included microphone works fine, while Segien thinks people would be better off buying a microphone, one that includes an on-off switch. I used the packaged microphone. (One tip: Turn the mike off when you answer the phone. Sigh.)
One of the first things I noticed is that I no longer had to speak one ... word ... at ... a ... time. For the best results, I had to speak in a conversational tone with as few breaks as possible.
Dragon NaturallySpeaking claims that this version is 99 percent accurate. Eventually, that may be true for me, but not right out of the box. I found myself raising my voice when the screen showed something far different from what I said, and attempts to correct the errors failed.
I am not the only one who has had difficulty with speech recognition lately. At a demonstration last month of the speech recognition capabilities in Windows Vista for financial analysts, a Microsoft staffer said "dear Mom," but it came out "dear aunt." When he said "fix aunt," it came out "let's set." He unsuccessfully tried to delete it three times.
That was similar to my experience, though I found making the corrections more responsive than I recall from my other tests. A lot of it has to do with both the user and the software learning each other.
"The earlier versions required users to adjust more things to get it to work," Roach said. "You had to have the microphone quality good. You had to have a computer with a very good sound card in it. It had to be a head-mounted microphone."
On a lighter note, Roach said people who chew tobacco have more difficulty with speech recognition. Why? Because they don't move their jaws enough.
The software comes with a quick start guide that lists the most common commands. It also has a good users manual, which I referred to frequently. Starting out, I probably would recommend creating your own guide until you have the commands mastered.
While the dictation was probably 80 percent accurate, one improvement jumped out. I navigated the desktop and other applications easily. I opened Microsoft Word, Mozilla Firefox and other programs without touching the keyboard or mouse. I surfed the Web without any problem, and I moved the mouse with voice commands.
However, the voice is not as speedy as the hand, and I was often tempted just to grab the mouse and click, just as I did with corrections and rewrites on this column. "It's not the computer," Roach said. "It's the people. When we conversationally talk to somebody, our voice is so uneven, even though we don't realize it."
The software mastered one of our traditional tests almost immediately. "The rain in Spain" from My Fair Lady came out almost perfect the first time. I just had to change "plane" to "plain," unlike some of the mangled versions in our reviews years ago.
Perhaps next time I should substitute the song from Mary Poppins, the one with the really long word that starts with "super." ... On second thought, let's leave well enough alone.
Speech recognition software lets users record their words by speaking into a microphone instead of typing on a keyboard. A computer program matches one's speech with a database of recognizable words and phrases.
The software has improved over the years, but it's not perfect right out of the box.
Some examples, based on my test:
|IT TYPED |
|"Hello, Ed" || The low end |
|"I've tried" || Five try|
|"At" || Acts|
|"Getting" || Gathering|
Information from Times wires was used in this report. Dave Gussow can be reached at (727) 445-4165, or e-mail firstname.lastname@example.org Read his blog at www.sptimes.com/blogs/tech.
[Last modified August 29, 2006, 16:55:34]
[an error occurred while processing this directive]
Subscribe to the Times
for daily delivery
of the St. Petersburg Times.