Tired of typing? Try talking to your computer

While no one is quite ready to throw away the keyboard, some workers are finding a useful alternative to typing: All that's needed is a late-model computer, special software, and a microphone.

Since practical desktop speech- recognition software first appeared in 1995, it has promised to type words faster than anyone's fingers could fly.

But technical glitches severely limited its appeal. Total retail sales of voice recognition software are only about 5 percent that of virus protection software, according to the NPD Group, a market research firm in Reston, Va.

But thanks to steadily improved software running on the latest generation of PCs, speech recognition is beginning to live up to its hype, spitting out 120 to 160 words per minute (w.p.m.) - the speed at which most people talk. In certain professional settings, that speed can prove invaluable. Some doctors, for example, no longer need transcription services.

When David Heiman, a gastroenterologist in Tampa, Fla., speaks into his computer, his words quickly appear on screen. He then edits the memo and sends it out to the referring doctor the same day. Colleagues who use transcription services, he says, often have to wait at least a week before their memos are finished.

"Doctors have stopped me in the hall to ask how I get such fast turnaround," says Dr. Heiman, adding that he saves about $1,000 monthly by not paying for transcription services.

But the technology does have a downside - accuracy. While all the words generated by the system are spelled correctly, they may be the wrong words.

"Speech-recognition software keeps getting better, but in the meantime it is still frustrating," says Paul Kiesel, an attorney who uses Dragon NaturallySpeaking speech-recognition software from ScanSoft Inc. at his office in Beverly Hills, Calif.

Mr. Kiesel says that while six other attorneys in his firm use the software, another six got frustrated and went back to their keyboards. Even after eight years of practice, Kiesel figures he achieves about 90 percent accuracy, at his personal dictation rate of about 100 w.p.m.. Assuming 200 words on a double-spaced page, that means 20 corrections per page.

"But if you are speaking at 100 w.p.m. and getting 90 percent of them down, you are still way ahead of someone typing 20 w.p.m.," notes Amy Wohl, an office-automation expert and head of Wohl Associates in Narberth, Pa. (Ms. Wohl doesn't use the software since she types at 100 w.p.m.)

Vendors claim speech-recognition devices have up to 99 percent accuracy in controlled settings. And Kiesel says he has gotten such results when reading printed material.

Correcting mistakes is usually a matter of highlighting the incorrect word and pronouncing it again, and the systems can be "trained" on troublesome or uncommon words. Long, idiosyncratic words (like "idiosyncratic") are more likely to be recognized successfully than short ones such as "he," "it," and "them." So it's no mystery that the medical and legal professions, with their cornucopia of big words, are growing users of these systems, notes Robert Weideman, chief marketing officer at ScanSoft Inc. in Peabody, Mass.

General business use is also on the upswing as employers seek to meet the needs of office workers with repetitive stress injuries said to be caused by typing, he adds.

ScanSoft claims the majority of the desktop speech-recognition market, with its Dragon NaturallySpeaking package. This is especially true after its largest competitor, IBM, recently sold the republishing rights to its ViaVoice speech-recognition product to ScanSoft. Prices for standard office packages range from $60 to $150. (Microsoft's Office XP suite also includes speech-recognition services.)

These desktop products are "speaker dependent," meaning that, out of the box, they have to be "trained" by the owner reading canned text into the microphone.

Airline-reservation systems and direct-sale catalog operations rely on "speaker independent" devices, meaning they can be used by nearly anyone. But the latter depend on fancier computers that focus on limited vocabularies. (After all, you're not going to call your airline to dictate a romance novel.)

As for using the systems, the best results require a fast computer - the latest 2 gigahertz PCs suffice - and digitizing headsets with USB plugs, says Bill Meisel, president of TMA Associates, a speech-industry consulting firm in Tarzana, Calif.

Mr. Meisel urges customers to let the speech software examine their existing word-processing files to ensure that their favorite words are in its dictionary. Then, use it for two weeks before deciding if it's right for you. During that time, you will get used to dictating, and the software will learn your pronunciation, Meisel says.