Look, no hands--but Voice 2.0 needs a lot of memory to work

Imagine controlling your computer or writing e-mail without a keyboard and mouse. Think
how it would reduce repetitive-stress injury and help users with disabilities.

If you're interested in voice recognition for any of these purposes, take a look at
Voice 2.0 for Windows. One caveat: You'll need a better-than-average desktop PC.

The minimum CPU here is a 75-MHz 486DX4; a Pentium is recommended for Microsoft Windows
95. You must have 8M of dedicated RAM to use the 30,000-word vocabulary and 16M for the
60,000-word vocabulary.

Fast Pentium computers with sound cards and 16M of RAM are becoming standard for new
computer buys, and 32M machines aren't uncommon. But this isn't the only hardware demand
for voice recognition. You must free up 35M on the hard drive and install a
Kurzweil-supported sound card--all SoundBlaster 16-compatible boards are supported, for
example.

Voice for Windows comes on floppy disks and CD-ROM. Installing the floppy version took
about 15 minutes. Except for selecting between the 30,000- and the 60,000-word vocabulary
options, installation was normal for a Windows program.

My PC has 16M RAM, so I chose the 60,000-word vocabulary option.

A Telex Nomad microphone supplied with the program required some assembly. It wasn't
difficult, but the instructions were a little ambiguous. I was eager to test the program,
so I started it as soon as I finished installation and received an error message:
insufficient vocabulary memory. Yet the box said 16M of RAM was enough, and my machine had
it.

After fiddling with the size of my hard-drive swap file and poking around in the
Control Panel, I called tech support at Kurzweil's toll-free number. Everyone was busy, so
I left a voice message. Then, awaiting a return call, I wandered around on Kurzweil's Web
page at http://www.kurz-ai.com  and
found my answer.

To run the 60,000-word vocabulary, I needed 16M of dedicated RAM, not
including the memory required for Windows and its applications. In other words, with
Windows 95 you must have 16M overall to use the 30,000-word vocabulary and 24M to use the
60,000-word version. In Windows 3.1, you might get by with 4M less, but more is better.

Tech support returned my phone call within two hours and confirmed my finding: I could
run only the 30,000-word vocabulary on my PC. So I reinstalled Voice for Windows that way
and had no problems.

The first time you use Voice for Windows, you speak your name and gender as a new user.
Putting in the gender increases the accuracy of recognition. After you speak about 800
words, Voice will prompt you to complete your enrollment by recording 480 words that
contain all the sounds in the English language and 33 number sequences for continuous
digit recognition.

This enrollment process is supposed to improve overall recognition accuracy to better
than 90 percent.

You can speak commands to control the mouse, keyboard, and individual application and
operating system functions. Kurzweil's vocabularies contain the menu commands of at least
30 popular programs, and there are easy provisions for'20adding new programs and menu
commands.

Switching between applications with Voice is a simple matter of saying "Open
(application name)." Keyboard and mouse control are less intuitive.

When you substitute voice control for the keyboard, words must be spelled a letter at a
time in the military phonetic alphabet. Although accurate, it's a slow way of spelling.

Control the mouse by saying a number that represents how many screen pixels you want to
move the pointer and its direction. After a lot of trial and error, you can get quite
accurate at guessing how many pixels to move the pointer.

I found it much more difficult to give dictation to a word processor or e-mail program.
Two problems conspired to frustrate my attempt at dictating this review into my word
processor.

First, Voice for Windows does not accept continuous speech when receiving dictation.
You must hesitate slightly--about one-sixth of a second--between words. Second, when an
error appears, you should correct it immediately to teach Voice for Windows your speech
patterns faster.

These two annoyances resulted in what I term "temporary creative memory
loss." I kept losing my train of thought when speaking in what to me is an unnatural
manner and being interrupted repeatedly to correct an entry.

There are two schools of thought on correcting voice recognition errors. One advises
you to complete your document and then fix errors. The other advocates correcting errors
as they occur. Kurzweil has chosen the second option.

When you notice an error, you can fix it immediately by choosing an option in the Take
box, a dialog box that displays the most likely alternatives to the word the software
recognized. If the word you want is one of these options, you simply say "Take
One," "Take Two" and so on.

If the right word doesn't appear in the Take box, you say "Correct that,"
causing a second dialog box to appear. Then you enter the correct spelling with the
keyboard or by voice.

When the correct word appears, you "take" it and then continue dictating.
Voice for Windows learns as you correct. Proper use of the Take and Correct boxes will
familiarize the program with your speech and eventually eliminate the recognition errors.

I found that Voice for Windows really shines as a "mouse eliminator" for
controlling programs. I use it to keep my hands on the keyboard without learning all the
Alt-whatever shortcuts.

As a dictation device, Voice could be equally useful if you can learn to insert the
slight hesitations between words and still retain a coherent thought.

Another Kurzweil product, called VoicePad for Windows, is downloadable from the World
Wide Web. It's scaled-down, $49.95 shareware with a 12,000-word dictionary, to which you
can add 500 more words. You must provide your own microphone or buy one from Kurzweil for
$19.95. The 5M program requires Windows 95 and a Pentium CPU.

William M. Frazier, a PC hobbyist, is the postmaster of Ocean Shores, Wash.

Kurzweil Applied Intelligence Inc.

Waltham, Mass.; tel. 800-380-1234

Street price: $493

Overall grade: B

[+] General-purpose voice recognition that actually works

[+] Controls menus without a mouse

[-] Heavy hardware requirements

[-] Error correction interrupts dictation

Real-life requirements: 486 or Pentium PC with 16M or more RAM, Windows 3.x in
enhanced mode or Windows 95, SoundBlaster 16 or other supported sound board, 35M free on
drive