Let your talking do the typing (NNN May 1, 1999)

Einstein only learned to speak at age three.

So much for our preconceived notions about speech, intelligence and learning. I discarded many such fixed ideas when, 25 years ago, I read Peter Farb's "Word Play: What happens when people talk." It's still in print (ISBN 0679734082), and I commend it to you.

I've had to discard another slew of speech prejudices recently, when I encountered IBM's ViaVoice 98 Office. You've seen the commercials: "You talk, it types."

It works.

I have learned to have low expectations for software, especially for complex applications that suggest they're going to change my working life. Such programs often do result in changes, usually by making me invent new invectives to inveigh against the poor, dumb machine that suffers installation of over-hyped, under-performing, cute but contrived applications of poorly conceived, incompletely developed toys.

Given the inherent confusion of the spoken word -- variations in pronunciation, spelling, meaning, clarity and context -- and the limitations of machinery, speech recognition technology has a long way to go before we can hold meaningful conversations with computers.

But if one has need of the facility to dictate documents to a PC, rather than type them, ViaVoice 98 demonstrates that it can be done relatively smoothly.

Setup requires training the program to recognize your voice, pronunciation and speech style. I spent about 1.5 hours (skeptically) reading to it, from sentences and short stories presented on the screen. I then watched the short video tours that explain its workings.

Then I said, "Dictate to SpeakPad" (the bundled word processor, which promptly opened), and began speaking in natural sentences. Everything I said was typed on the screen.

It worked best when I spoke fluently and unhesitatingly, worst when I tried to make allowances for the program by speaking in a stilted style, as if conversing with a child or someone not fluent in English. But I soon ran out of things to dictate spontaneously. So I picked up a handy financial report and read from its esoteric descriptions of global economic transactions. That worked even better.

There are three versions of ViaVoice 98: Home (about $40-50 depending on where you buy it) includes a built-in vocabulary module called "Cuisine;" Office (the one I tested, about $61-90) includes a business/finance module; and Executive ($110-150), which has some extras.

There are additional vocabulary modules, called "Topics," available separately, about $10-20 each. You can buy extensive vocabularies for the legal and medical professions ($150 each), the latter also available in versions for various medical specialties.

You can add to the built-in vocabularies yourself by reading the words in, and ViaVoice 98 is available in other languages. Online Companion, a $12-15 add-on, voice-enables some e-mail and chat applications, including Netscape, Internet Explorer and AOL.

(IBM has also released a talking Web browser designed for the blind, about $149.)

Minimum requirements are Windows 9x/NT, sound card, fast Pentium (166MHz or better) and at least 32mb RAM. I used an AMD 300 with 64mb. A headset microphone is included.

If you allow its language blueprint sufficient time to take on the shape of your speech, and are patient enough to teach it as you go, you may find your biggest problem is getting tongue-tied.

You talk, it types. Be careful what you say.

E-mail: NNN

Jerry Maizell

NNN
Near North News
Chicago