This has great potential for students and adults with many sorts
of disabilities...


Researchers Bring Voice Recognition to Palmtops

October 11, 2001

By LISA GUERNSEY




OVER the past few years, Liam D. Comerford, a speech
engineer at I.B.M.'s Thomas J. Watson Research Center, has
developed a habit of conversing with his Palm III.
Sometimes he tells the device about his chores. Other times
he will listen as the machine reminds him of a scheduled
meeting. Once in a while he'll simply ask it for the time.

Mr. Comerford is not losing his mind. His organizer can
actually recognize his speech in addition to uttering
sentences itself. It is one of nearly 100 Palm organizers
that I.B.M. (news/quote) Research has reconfigured in an
attempt to create a speech system small enough to reside on
a hand-held computer.

"I'm ready now, Liam," his Palm announced in robotic
staccato during a recent demonstration.

"How do you feel?" its owner asked.

"My battery power is
at 83 percent," the device responded.

Five years ago, hand-held devices with speech interfaces
were futuristic visions. Someday, engineers predicted,
people would get so fed up with tapping on tiny keyboards
that they would want to talk to their devices instead. Some
people, they figured, would find small screens such a
strain that they would want their hand-held computers to
read their notes aloud, particularly while they were
driving.

But getting speech systems into small devices means
overcoming size and power constraints. Until recently,
hand-held computers were equipped with processors with only
limited power to avoid draining the batteries too rapidly.

Speech technology was out of the question because of the
processing power it required. Indeed, several companies
decided to forgo the idea of embedded systems in favor of
devices that can generate or recognize speech only when
connected to a larger server or computer network. Microsoft
(news/quote), for example, has developed technology for
hand- held computers that have a wireless connection to
remote servers.

Yet as processors and batteries have become more efficient,
some companies have held on to the idea of putting entire
speech engines within hand-held computers. They are
convinced that people will become frustrated with hand-held
speech products that require wireless communication to
operate, particularly given the lag time and access
problems. They want to offer systems that are entirely
embedded in devices so that they will always be ready for
use.

The system demonstrated by Mr. Comerford is one example.
I.B.M. Research has been working on the technology, which
it calls the personal speech assistant, for several years.
He said that the protoype was based on three decades of
work by I.B.M. scientists to create increasingly nimble
programming codes for speech systems. Faster and smaller
processors also helped, he said, as did a few improvements
in hardware, like adding a speaker, a microphone and an
additional processor to the Palm.

The technology can also be adapted for other hand-held
devices. Last week, in fact, I.B.M. announced that part of
its speech system had been adopted by Compaq for use in its
new iPaq 3800 series.

Consumers might soon get another taste of a talking
hand-held device from Lernout & Hauspie, the speech
software company in Burlington, Mass. In the next few
months the company plans to release software called PDSay,
said Pam Ravesi, senior director of product management. For
about $35, she said, people will be able to download the
software into any Compaq iPaq, models 3630 and later.

SpeechWorks, a software company in Boston, has also
developed a few embedded speech systems for hand- held
computers, said Mike Phillips, the company's chief
technology officer. The systems can turn text into speech
and can understand spoken commands like those used to open
files or read addresses, he said. Mr. Phillips would not
say which manufacturers would be working with the company.

Another forthcoming speech system was demonstrated last
month in San Diego at DemoMobile, a technology conference
for investors and technology executives. Developed by Voice
Signal, a speech company in Woburn, Mass., it is called
Elvis, an acronym for "embedded large vocabulary interface
system." Unlike the I.B.M. prototype, Elvis does not
generate speech but it does recognize voice commands and
can take dictation. It is expected to go on the market for
manufacturers by the end of the year.

"The challenge was, how do we make a dictation package that
fits in 1 megabyte of memory instead of 128 megs of
memory?" said Dan Roth, Voice Signal's chief executive.
"And we've been able to do it."

The breakthrough was achieved partly by compressing the
system's accompanying dictionary and re-engineering some of
the software code and the core speech engine for use on
small devices, Mr. Roth said.

At the demonstration, a Voice Signal official dictated a
100-word e- mail message that was captured and translated
into text on a Compaq iPaq. Mr. Roth said he expected this
type of interface to be popular with people who are
currently struggling with tiny keyboards and styluses to
tap out e-mail messages in cabs on their way to meetings.

I.B.M.'s technology could be used for e-mail, too. But so
far, Mr. Comerford has only been testing its ability to
dictate appointments and memos to save for his own viewing
- or listening - in the future.

For example, he has programmed the device to alert him to
meetings. "You have an appointment at 2 p.m.," it tells
him. Mr. Comerford has also found the speech assistant to
be helpful in recording expenses: when he gets out of a
cab, he said, he tells the device how much the ride cost.

The concept of a talking device does have its drawbacks.
Imagine, for example, that a person's Palm started giving
away secrets, perhaps by making an announcement about a
confidential meeting that could be heard by anyone within
earshot. And what about the annoyance of hearing robotic
babble every few minutes?

I.B.M. has considered those problems, Mr. Comerford said,
and is working to make it easy for people to move
seamlessly from a speech-oriented system to the stylus and
back again. And it is trying to keep the Palm mute when
necessary.

"Right now, we want it to ask, `Hello, are you there?' "
Mr. Comerford said. "If it doesn't get a reply, it should
put itself to sleep."

Such niceties, however, will not override another potential
problem. With hand-held devices taking dictation, train
passengers may have to put up not only with cellphone
talkers but also with the drone of passengers dictating
their to-do lists.

http://www.nytimes.com/2001/10/11/technology/circuits/11VOIC.html?ex=1003831
664&ei=1&en=6db5883aebb94c9a



HOW TO ADVERTISE
---------------------------------
For information on advertising in e-mail newsletters
or other creative advertising opportunities with The
New York Times on the Web, please contact Alyson
Racer at [log in to unmask] or visit our online media
kit at http://www.nytimes.com/adinfo

For general information about NYTimes.com, write to
[log in to unmask]

Copyright 2001 The New York Times Company