Voice recognition package says yes to 'Can we talk?'
Consider me impressed. It isn't perfect, but NaturallySpeaking is an advance over
earlier speech recognition software. It took 30 minutes of reading a Dave Barry book aloud
to train it to understand my voice.Some other recognition applications take as long as
three hours to enroll, and their chosen text is less entertaining.
NaturallySpeaking keeps learning as you talk. Within moments of completing the
30-minute training, I began dictating this review. For the most part, the software
The first problem was with my name. When I said "Michael Cheek," it heard
"Michael G." To give it a spelling lesson, I said "Select G," and the
letter G was highlighted. Then I said, "Spell that." A dialog box appeared with
My last name was not among them. So I spelled out loud: "Cap C-H-E-E-K click
OK." The next time I spoke my name, NaturallySpeaking got it right.
The Spell That command is handy. If the word you're saying is among the 10 choices that
appear, simply say "Choose" followed by the number. Or, as with my name, spell
the word out loud.
Other handy commands are "Scratch that" to delete your last spoken phrase,
"New paragraph" and "New line" to start a new paragraph or line,
"Go to" or "Move" to navigate in the document, and the powerful
"Select" to highlight any word, line, phrase, paragraph or document.
You must specify the exact punctuation, which can slow things down. Although the
package is called NaturallySpeaking, natural speech will cause some words to be lost or
misinterpreted. But you don't have to say words with pauses between. The best cadence is
like that of reading aloud.
Don't forget to enunciate. I grew up in Georgia and tend to drop the last letters of
words. But when I took care to enunciate, more words came out correctly.
NaturallySpeaking can be frustrating, especially while it learns. There are occasional
pauses as the application caches information about your speech patterns to the hard drive.
I tested it on a 200-MHz Pentium PC with Microsoft Windows 95 and 32M RAM, and I would
consider that a minimum for acceptable performance.
At $299, you can't go wrong with this package, which includes a headset and microphone.
Given a little time and clear pronunciation, you and NaturallySpeaking will understand
each other quite well.
Here's a sample from my dictation of the beginning of Abraham Lincoln's Gettysburg
Address. NaturallySpeaking errors appear in italics and my typed corrections in brackets:
"For [four] score and seven years ago, offers [our fathers] brought forth on this
[continent] and [a] new nation, conceived in liberty and dedicated to the proposition that
all men [are] created equal. Now we are engaged and [in] a great civil war, testing
whether that nation were [or] any nation so conceived and so dedicated and [can] long into
"We are met on a great battlefield of the[that] war. We have come to dedicate [a]
portion of the [that] field as a final resting place for those who hear [here] gave their
lives that that nation might let [live]. It is altogether fitting improper [and proper]
that we should do this. But in a larger sense, we cannot dedicate, we cannot concentrate
[consecrate], we cannot howl of [hallow] this ground."
NaturallySpeaking came close to the actual text with an error rate of about 14 percent
with similar-sounding words. Compared to earlier speech recognition software,
NaturallySpeaking is a leap forward.