Computers Still Have A Tin Ear

Captain Jean-Luc Picard doesn't have to know how to type. When the commander of the starship Enterprise wants something from his computer, he just asks for it.

For many of us stuck in the 20th century, however, the keyboard remains a barrier to efficient use of computers. Keyboard-phobia is widespread, especially among middle-aged men who got into business thinking that typing was something their secretaries did. That creates a big potential market for computerized speech recognition. Despite enormous improvements in technology and some useful specialized applications, the day when you can easily and naturally dictate a letter or memo to your computer remains well in the future.

The adventurous can get a glimpse of that brave new world with software-hardware combinations now on the market. The development of fast, cheap digital signal-processing chips has helped speech recognition grow from laboratory toy to useful tool. But practical applications are still limited and specialized. Telephone companies are using the technology to replace human operators, and airline reservations offices and catalog-sales operations are developing their own systems. These highly structured situations typically involve simple commands and conversations. Says Robert W. Lucky, vice-president of Bellcore, the research arm of the regional Bell telephone companies: "The sort of normal conversation we're having just is not going to happen" with computers.

Kurzweil Applied Intelligence's new Voice for Windows is a pricey but comprehensive introduction to speech recognition. The $995 package lets you use speech to replace keyboard commands or mouse clicks in a number of popular programs. In addition to these "command and control" functions, it--and Dragon System's similar Dragon Dictate for DOS--let you dictate text to your word processor or numbers to your spreadsheet. Sort of. Voice uses what's called "discrete" speech recognition--it can only understand language one word at a time. This requires you to speak with a distinct pause between words, a procedure I found difficult and unnatural.

Speech Systems' Phonetic Engine takes a very different approach. It lets software developers write programs that understand "continuous" speech, meaning you can pace your words naturally and the computer will parse the sentence. Of course, there's a catch. The system is limited to a narrow vocabulary and tightly controlled syntax. For example, the company used its $995 development system to write a program designed to train air traffic controllers. It's good at understanding statements such as "Delta 5-9-2 clear to land on runway 2-7-L" but can't handle "Please get me a cup of coffee." The difference is that in controller-speak, the software can count on familiar patterns; for example, an airline name is always followed by a number.

CUT AND PASTE. Even crude speech recognition can be a godsend for people with disabilities. And it can be very helpful on jobs where you can't spare your hands for the keyboard. For example, speech-recognition systems are being used to help doctors record their actions in the emergency room or operating room.

Broader use awaits breakthroughs in artificial intelligence that will allow computers to understand natural, spoken language. And programs will have to be rewritten to make them easier to use with spoken commands. For example, deleting a sentence in Microsoft Word using the Kurzweil product requires a string of commands, such as: "Edit. Select Current Sentence. Edit. Clear." That's because the program is designed to be controlled by a mouse and the keyboard, not by spoken commands.

So don't expect voice-controlled programs to become a hot item at your local Egghead Software store anytime soon. As Bellcore's Lucky puts it: "It's still a lot easier to type certain things than to say them."

Before it's here, it's on the Bloomberg Terminal.