`Computer, Save This File'
Just a few weeks ago, I wrote that two new dictation products solve half the problem of using speech to run a computer (BW--Sept. 22). Dragon Systems NaturallySpeaking and IBM ViaVoice both do a good job of turning the spoken word into text. A new product from Kurzweil Artificial Intelligence takes on the other half of the problem by letting you accurately and efficiently control your computer with speech.
We're not quite at the promised land yet. Kurzweil's VoiceCommands (Alpha Software, 781 229-2924) works only with Microsoft Word 97 or 7.0. And trying to integrate it with a dictation program is a difficult do-it-yourself job. But Kurzweil, a subsidiary of Lernout & Hauspie Speech Products, hopes to ship an integrated dictation-and-command program by yearend. Dragon and IBM are racing to add command-and-control features to their products.
AU NATUREL. The breakthrough in VoiceCommands, available for around $70 including an Andrea Electronics noise-canceling mike, isn't its ability to understand spoken commands: Software that can execute precise spoken menu commands has been around for a while. But VoiceCommands allows you to tell Word what you want to do using what artificial-intelligence experts call "natural language."
For example, suppose I have this column open in Word and say: "Select third paragraph. Bold it. Italicize it. Center it." Almost as quickly as I speak, and without any reference to Word's complicated menu trees, the paragraph above this one will be reformatted in centered, bold italics.
The ability to comprehend natural-language commands is no great trick. Because the vocabulary of commands is limited, the program has to understand only a few dozen words, compared with 20,000 or more for an effective dictation system. The limited vocabulary also means you can start to use the program without the tedious speech training required by dictation systems.
VoiceCommands can handle all of the most commonly used functions of Word and a few of the more obscure ones, too. For example, tell it to "turn the next five lines into a table," and it will do just that. To use commands that the program doesn't know, you can talk your way through menu trees and dialogue boxes.
I found the program was largely free of glitches, though occasionally it would fail to understand a command, and I'd have to repeat it. I couldn't make it understand the command "backspace," and a Kurzweil engineer confirmed that the problem appeared to be a bug.
COMBO. Of course, what we really need is a system that combines dictation with voice control--and works with all the programs on your computer. Making such software function both smoothly and consistently requires that the operating system itself support speech recognition. Apple Computer Mac OS included a version of speech command-and-control called PlainTalk for some time, and IBM's OS/2 Warp added the feature last year. But neither comes close to VoiceCommands in flexibility and ease of use.
Microsoft, despite extensive speech research, has proved a laggard at incorporating the technology into its products. However, the software giant recently took a $45 million equity position in Lernout & Hauspie and announced a "strategic relationship" to incorporate speech technology into its products. A version of Windows that handles spoken commands like VoiceCommands and also takes dictation with the facility of ViaVoice or NaturallySpeaking would make hands-free computing a reality.