IBM Unveils Low Cost PC Run by Voice

International Business Machines lent its considerable clout to the growing market for speech-recognizing personal computers on Thursday by introducing a low-cost system that recognizes up to 7,000 spoken words.

IBM said its new voice-recognizing system will allow thousands of disabled people, many of whom are now unable to type on conventional keyboards, to use computers.

Although IBM VoiceType is not the first-ever speech-recognition computer system, its cost of about $7,200 makes it the most affordable, a breakthrough that could help it reach a market far beyond the handicapped users for whom it was initially developed.

Designed with Dragon Systems of Newton, Mass., the new system--consisting of a computer circuit board and software--allows a user to dictate words or numbers into a personal computer for word processing, data processing, spreadsheet calculations or other applications. After listening to slowly spoken words or instructions, the system “types” them onto the computer screen. The system also has a memory feature that enables it to type up to 1,000 preprogrammed keystrokes with a single verbal command.

Although the system comes with a built-in dictionary of about 80,000 words, it can be programmed to understand only a vocabulary of 7,000 words at a time. And unlike the talking computer Hal in the science-fiction movie “2001: A Space Odyssey,” IBM’s VoiceType does not itself have a voice, so the system does not talk back to its user.

“We’re hoping this opens a whole new job spectrum for people who have been blocked from other paths,” said Walt Nawrocki, product manager for IBM’s special needs systems group in Boca Raton, Fla. The system is expected to be available in August.

While the IBM system represents an improvement over others now available, analysts say voice-recognition computers are still far from achieving their original and much-predicted promise of understanding fluid, conversational speech.

“These systems are not better than typing, if that is available to you,” said Richard Shaffer, a technology newsletter publisher in New York. “They are not accurate enough,” and they still don’t understand a wide enough vocabulary to be an alternative for the everyday computer user.”

However, researchers at AT&T;’s Bell Laboratories in New Jersey say breakthroughs are on the horizon to bring speech recognition into everyday use before the end of the century.

“Right now the issue is cost,” a Bell Labs spokesman said. “But we believe that by the year 2000, computers will understand continuous speech and fluid sentences, and that we will be able to fit the capacity to recognize a vocabulary of 20,000 words on a single computer chip.”