Interesting article from today's Boston Globe. Maybe we will ge a
mention someday.
From the Bloston Globe 6/8/99:
[ Send this story to a friend | Easy-print version | Add to Daily User ]
Hello, future
While mass market for speech-recognition software has been slow to
materialize, firms are rushing to embrace technology
By Hiawatha Bray, Globe Staff, 06/08/99
ave you talked to your computer lately?
Probably not, if you're an average computer user. Although there's plenty of
software that lets computers understand the human voice, most people stick to their
old-fashioned keyboards.
But if you don't talk to your own computer, you may talk to somebody else's. Many
businesses have phone systems that'll connect you to the right person if you just say his
name. Some retail businesses, such as brokerage houses, will let you buy and sell stocks
with your lips instead of your fingers.
While the average person may not use speech-enabled computers today, busy professionals
such as doctors, attorneys, and even police officers are embracing them as a fast, reliable
way to stay on top of their paperwork.
And as computers become ever smaller and more powerful, speech control will be included
in even the most ordinary devices. Already there is a car radio that can be controlled by a
driver's voice. Someday home TV sets may work the same way - along with every other
appliance in the house.
As experts in computer speech recognition gather in Boston for an industry conference
today, it's clear that speech recognition is catching on in a big way - just not the way many
people expected.
''Any new technology that you look at actually takes a while to take off,'' says Roger
Matus, vice president of Dragon Systems Inc., one of several Boston-area firms at the
forefront of speech technology.
That's been the pattern so far with speech-recognition software. In recent years, companies
including Dragon, IBM Corp., and Burlington's Lernout & Hauspie have sold
speech-recognition software at retail. With computers in half the homes in America and
nearly every office, there would seem to be a vast market for such software among users
who've never learned to type.
But so far, a mass market for these products has yet to appear. According to market
research firm PC Data Inc., speech-recognition software makers sold just 775,000 units at
retail last year, worth $59.7 million.
But at the same time, companies are rushing to deploy speech-recognition systems that
work in conjunction with telephone systems. Industry-watcher William Meisel of TMA
Associates in Tarzana, Calif., estimates that more than $2 billion worth of such software
will be sold this year, and more than $36 billion by the year 2003.
William ''Ozzie'' Osbourne, manager of IBM's speech systems group, says his company
is investing heavily in speech-controlled telephone products. Osbourne uses one every day
- an internal phone directory. He can dial IBMNAME and say the name of any other IBM
employee, anywhere in the company. The computer recognizes the name, looks up the
number, and completes the call.
Parlance Corp. of Medford has sold similar systems to about 100 companies, including The
Boston Globe.
''People who are traveling love it,'' said sales executive Susan Richards, ''because they
can reach any colleague in the company by dialing a single number.''
Boston-based SpeechWorks Inc. produces more advanced speech systems used by
companies that sell by phone. For instance, the Internet brokerage firm ETrade uses a
SpeechWorks system to let customers make stock trades from any phone, without ever
speaking to a human being.
Paul Lightfoot, president of New York-based Foodline.com, is using a SpeechWorks
system to offer a telephone guide to New York restaurants.
''If the service isn't fast and easy to use,'' says Lightfoot, ''it's not going to be used.''
So he rejected a system that might require a user to press 1 for Italian food and 2 for
French. Instead, the user just says ''I want Italian food,'' and the software responds with a
list of choices.
Telephone-based speech software has to be able to recognize many different voices and
accents. On the other hand, most of these systems are designed for a specific purpose, so
they need to know only a few hundred words.
Ironically, this same limitation applies to some very highly technical activities, including
medicine. The leading speech-software firms have done quite well with customized
software for use by doctors.
At Quincy Hospital, chief of emergency services Octavio Diaz and a staff of nine doctors
write up about 90 percent of patient reports using medical transcription software sold by
Lernout & Hauspie.
''It's a very user-friendly system once you get up and running on it,'' said Diaz, who has
been using speech-recognition software for about 10 years.
At Massachusetts General Hospital, clinical cardiologist John Le-
vinson swears by his Dragon software.
''Filling out a visit form is like filling out a 1040 long form for each patient,'' says
Levinson. So he dictates patient information instead of typing. ''Most doctors who have
come through my office and seen me use my system immediately want it.''
Having found a stable market among doctors, firms such as Lernout & Hauspie and
Dragon are also tailoring custom speech-recognition products for attorneys and even police
officers who hate filling out paperwork.
Now, the makers of speech-recognition software are looking for ways to make their
technology indispensable to corporate America.
Lernout & Hauspie's parent firm is based in Belgium, and many of its engineers speak
several languages. No surprise then that chief executive Gaston Bastiaens is placing a big
bet on a difficult problem - computers that not only recognize spoken words, but instantly
translate them from one language to another.
Lernout & Hauspie says it's well on the way to solving this problem. The company
recently showed off a product called Global Chat that instantly translated the words of a
speaker from English to German or another language. Not only does Global Chat display
the translation on a computer screen, but the computer actually pronounces the words, in a
surprisingly lifelike voice.
''This is something no one else in the world can do except us,'' Bastiaens boasts. Global
Chat, which requires corporate-grade computing power, is set to go on sale this fall.
Not to be outdone, Dragon is hard at work on an impressive new trick of its own called
''audio mining.'' This software analyzes recordings of human speech, identifies each
word, and creates an index. Thousands of firms record customer service calls to check up
on workers.
But with audio mining, there'll be no need to actually listen to the calls. Instead, the
computer can detect when Operator 25 calls a customer ''stupid.''
Ultimately, when powerful computers become cheap enough, the goal is to enable
practically any electronic device to recognize spoken words: a cell phone, for instance, or
personal digital assistants - PDAs - such as the 3Com Palm Pilot.
''It's going to work extremely well in PDAs,'' says IBM's Osbourne, ''because a
keyboard is not practical.''
Even some larger devices will eventually be well suited to speech control. Osbourne
imagines a time when you'll never again have to wrestle with setting your VCR. You'll just
tell it to record ''ER,'' and it will.
''We're not there yet,'' says Osbourne. ''But this is the future.''
|
