VICUG-L Archives

Visually Impaired Computer Users' Group List

VICUG-L@LISTSERV.ICORS.ORG

Options: Use Forum View

Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Kelly Pierce <[log in to unmask]>
Reply To:
Kelly Pierce <[log in to unmask]>
Date:
Sun, 2 Jul 2000 10:19:34 -0500
Content-Type:
TEXT/PLAIN
Parts/Attachments:
TEXT/PLAIN (308 lines)
All links in this article are numbered and appear at the end.  This
direction is exciting.  It brings web access to a new level.

kelly




the Wall Street Journal
                               June 30, 2000

Page One Feature

The Hot New Thing Is the Old Phone
In Gimmick-Hungry Dot-Com World

   By NICOLE HARRIS
   Staff Reporter of THE WALL STREET JOURNAL

   In a tiny makeshift studio, Christopher "Blade" Kotelly is coaching a
   professional voice named Tom Glynn. "OK, you're sitting in a '50s
   diner. It's upbeat, but people want their cheeseburgers in a hurry,"
   says Mr. Kotelly, a 28-year-old staffer at Boston-based SpeechWorks
   International Inc. (www.speechworks.com1) "You want your voice to be
   quick, to-the-point, but you don't want to be patronizing."

   Mr. Glynn takes his cue and begins in his smooth delivery, "Welcome to
   MapQuest by phone." Mr. Kotelly sweeps the air with his hands as
   though conducting an orchestra. "Good," he says when Mr. Glynn
   finishes. "But I want you to change the tone of your voice on the word
   'quest.' Be cool and casual." Mr. Glynn takes a swig of water, clears
   his throat and launches into another take.

    [Go] 2SpeechWorks Plans Two New Products For Speech-Driven Web, Phone
                                                       Services (June 29)

     [Go] 3Flooz.com to Use Technology From Voice Firm NetByTel (June 28)

   [Go] 4IBM Forms Deals With Three Companies To Develop Voice E-Commerce
                                                       Platform (June 23)

    [Go] 5WorldCom, PhoneRun Announce Voice-Internet Marketing Deal (June
                                                                      20)

   At the cutting edge of the Internet, this is the race of the moment --
   building a service that untethers the Web from the PC and brings a
   vast new market online to shop, trade stocks, find a recipe or, in the
   case of MapQuest.com Inc.'s plan, locate and get directions to
   anywhere in the U.S. All that people will need is a telephone and a
   voice.

   With many of the nation's 90 million cell-phone customers, along with
   rising numbers of Palm Pilot devotees, using wireless Web technology,
   dot-coms have been rushing to make their content wireless-friendly.
   Some, including MapQuest, have teamed up with wireless carriers to put
   their content on cell-phone screens. Many more, thinking that only
   techies will want to bang out e-mail on cramped cell-phone pads, are
   also betting on voice-recognition technology that can be used on
   phones of all kinds -- putting them within arm's reach of nearly every
   consumer in the nation, computer-savvy or not.

   "As a dot-com, our reach was limited to those who got to the Internet
   through some type of techie device," says Kathy Kinney, MapQuest's
   director of business development, who has been overseeing the New
   York-based company's drive to provide its Web content by phone. "We
   looked at the phone service and said, 'Holy cow, this is the way we
   can reach the neighbor across the street through the phone in her
   kitchen that she's been using all her life.' "

   Already, America Online Inc. has had a hit with its MovieFone service,
   and E*Trade Group Inc. has offered stock-trading by voice for more
   than two years. More recently, so-called voice portals like Quack.com
   (www.quack.com6) have been providing speech-recognition packages to
   the likes of Lycos Inc. and Bid.com International Inc., a
   business-to-business auction site for everything from heavy machinery
   to airplane parts. AT&T Corp. recently invested $60 million in TellMe
   Networks Inc. (www.tellme.com7), Mountain View, Calif., whose 1-800
   service provides stock quotes, traffic updates and news headlines, and
   may soon offer theater tickets and restaurant reservations.

   But transforming the Internet into a little voice in everyone's ear is
   a hard, slow slog. Witness the recording session at SpeechWorks'
   offices in Boston: Ultimately, the MapQuest system will require the
   recording of roughly 4,500 speech prompts, all tapping the
   old-fashioned talents of people like Mr. Glynn. Voice-recognition
   software is far from perfect, too, and computer generated voices often
   sound like just that-computer generated.

   Even with the glitches fixed, giving voice to the Web will require
   intensive re-engineering, including rewriting software code to
   translate text into speech. Service providers must either create
   applications in a new technological standard or pay for recorded
   content. Making one Web site wireless can run into hundreds of
   millions of dollars.

   It still has some of the kinks to work out, but MapQuest.com, which
   recently agreed to be acquired by America Online, will begin testing
   its toll-free "1-800 MapQuest" service in Denver in early July, and
   then roll it out nationally at the end of summer.

   Making the service free, MapQuest has decided to use audio ads to
   generate revenue. While the computer is translating a request and
   retrieving the directions to play back, the caller will hear a paid
   ad. In addition, callers may hear traffic updates that are "brought to
   you by" a particular sponsor.
   Kathy Kinney illustration

   Ms. Kinney says participants in early focus groups said they didn't
   want to be bothered by ads. But, she adds, after they were told the
   ads kept the service toll-free, they decided a brief commercial
   wouldn't be too "intrusive."

   The company seems like a natural for the wireless Web. Already, people
   sitting at their computers download about 10 million maps from the
   site daily, and many of them print out their maps to carry with them.
   MapQuest also licenses its technology to more than 1,500 businesses,
   mostly other Web sites. According to Media Metrix, MapQuest received
   nearly six million separate visitors in May.

   Ms. Kinney says MapQuest's services would be "logical" for mobile
   users who don't want to be tethered to a PC or who have no experience
   with them. She recalls one trial user of MapQuest's voice-recognition
   service saying that she would love to use the service to get
   directions instead of waiting for her husband to "fiddle with the
   wrong maps."

   The 42-year-old Ms. Kinney, trained as a cartographer, began
   orchestrating MapQuest's wireless Web strategy last year. Early on,
   she and her staff decided they couldn't go it alone and turned to
   SpeechWorks, founded by members of a Massachusetts Institute of
   Technology speech-technology group.

   Early focus-group tests offered some clues to the hurdles the partners
   faced: Most people would use the service on their cell phone while
   driving. Some people preferred to hear male voices, while others
   preferred female. One person wanted the service to disclose
   "historical sites, the cleanest rest areas and information about where
   the police hang out."

   MapQuest, too, had to adjust some of its ambitions to fit the
   technological reality. For example, some engineers wanted the service
   to let users listen to choices of cities and states when seeking
   directions, the way the Web site now offers lists of locales.
   SpeechWorks balked. As Roy Feldhusen, a SpeechWorks engineer, told the
   MapQuest team at the time: "God help you if you were looking for Zion,
   Kentucky." MapQuest deferred to SpeechWorks. Users will be asked to
   specify a city or state.

   Designed mostly for dictation, early forms of voice-recognition
   technology required users to speak very slowly, each word pronounced
   distinctly. The systems were difficult to use, limited in vocabulary
   and often didn't recognize words correctly. But speech-recognition
   software providers like SpeechWorks and Nuance Communications Inc.,
   Menlo Park, Calif., rely on the latest generation of the software,
   which thanks to increased processing power has much greater accuracy.
   This has allowed the companies to create large databases for, say,
   multiple pronunciations of the same word, so that a computer can
   recognize a Southern drawl as well as a Brooklyn honk.

   The MapQuest phone service is a combination of speech-recognition
   software and "text to speech" technology. A speech-recognition
   "engine" will translate callers' voices into digitized computer
   commands for mining the MapQuest database for directions. Then, to
   give the directions back to the caller, the system will use software
   that, in essence, translates the MapQuest Web content into audible
   speech -- either pre-recorded or computer-generated.

   Ms. Kinney says she originally wanted callers seeking driving
   directions to be able to say where they were calling from and where
   they wanted to go all in one breath. "The optimum for us would have
   been all-natural language ... having the callers being able to say
   exactly what they want at the outset, instead of being drilled down,"
   Ms. Kinney says.

   But that wasn't feasible. Requests for directions must be broken down
   into a sequence of ever-narrower prompts for parameters, from city and
   state, to street name and address, in order to sort through the
   database.

   Even then, from the experience of operating their Web site, MapQuest
   executives knew the speech system would have to be able to recognize
   even the most unusual caller response. Online users often typed in
   "BWI" for Baltimore-Washington Airport, for example, leaving it to the
   MapQuest computer to equate the two.

   Ms. Kinney wanted the phone service to allow for such discrepancies.
   "We also knew there would be multiple pronunciations of words, given
   people's accents," she says. Working with SpeechWorks programmers,
   MapQuest was able to create a database of hundreds of street names,
   cities, airports and other locations for the phone service, in various
   permutations.

   During the first round of testing (a focus group that included
   SpeechWorks and MapQuest employees), users discovered that New York's
   entire borough of Brooklyn was left out of the database. Then, after
   Brooklyn was added, the software didn't recognize it accurately. When
   a user spoke the words "Brooklyn, New York" after the computer asked
   for the city and state, the automated voice shot back: "I think you
   said Portland, New York. Is that correct?"

   In another test, when a user, responding to the prompt asking for an
   airport name, said "Dulles," the system couldn't respond because the
   database allowed only for the full phrase "Washington Dulles Airport."
   SpeechWorks fixes these glitches as they arise, and will continue to
   do so as users point them out.

   The "text-to-speech" part of the project proved more problematic.
   SpeechWorks is recording as much of the MapQuest content as it can
   using voice talent like Mr. Glynn for basic prompts such as "Where are
   you traveling from?" and "What's the name of the city and state?" But
   for much of the service, MapQuest had few options but to go with a
   technology still in its infancy. "There was no way in the world we
   could record every set of driving directions," Ms. Kinney says.

   One of the first systems SpeechWorks brought to her wasn't up to
   snuff. "I just shook my head," she says. SpeechWorks had better luck
   with text-to-speech software from industry pioneer Learnout & Hauspie
   Speech Products NV of Belgium.

   Still, such software overall "sounds like a drunken robot," says John
   Dalton, a consultant with Forrester Group. "It's tolerable for brief
   spurts of information, but you wouldn't want to hear a book or
   anything using this technology."

   SpeechWorks recently signed a deal with AT&T under which AT&T is
   licensing its text-to-speech technology to SpeechWorks in exchange for
   an undisclosed financial stake in the company. SpeechWorks says it may
   use the AT&T software, which sounds more natural and is very close to
   a recorded human voice, in the MapQuest package. (SpeechWorks itself
   recently registered for an initial public offering.)

   Once speech-recognition and text-to-speech software was in place, the
   fine-tuning began. Ms. Kinney in particular was worried about coming
   up with a way to allow users to retrieve step-by-step directions while
   driving. Focus-group tests had told her that callers would most likely
   use the service while driving, making it difficult for them to write
   down directions. But she didn't want users to have to stay on the line
   for long periods of time, running up costly 1-800 fees on MapQuest's
   dime.

   "Kathy wanted to make sure that people were encouraged and didn't feel
   nervous about getting off the line," says Steve Chambers, a
   SpeechWorks marketing executive.

   So SpeechWorks programmers have developed a "bookmark" system that
   allows users to pause the directions at any time. The voice prompt
   tells callers: "I'll reserve this place in the application and when
   you call back we can pick up from here." Hitting the redial button
   picks up where the directions left off after a few short steps.

   As the test launch has drawn closer, Ms. Kinney and Mr. Chambers have
   been overseeing the finishing touches on the human touch of MapQuest's
   new service. At an early meeting, Ms. Kinney and Mr. Chambers sketched
   a few raw ideas about the MapQuest voice that will be used to brand
   the service.

   "Maybe we can make it entertaining for users," Mr. Chambers said.

   Ms. Kinney asked for specifics.

   "Well, what do you think about the voice saying something like 'What?
   You're lost again,' " he said. "Or maybe if they choose Brooklyn, we
   can have a voice with a heavy Italian New Yorker accent, like the
   Sopranos."

   Even now, though, Ms. Kinney isn't sure whether she will use a male or
   female voice for the service. "There may be a question if a man will
   want to hear directions from a woman's voice," she says.

   At a recent focus-group gathering, the male option seems to be winning
   out. At least three different participants, male and female, request
   the voice of James Earl Jones. "He has the voice of someone I trust,"
   says one. "Someone who wouldn't get me lost."

   Write to Nicole Harris at [log in to unmask]
     _________________________________________________________________

   URL for this Article:
   http://interactive.wsj.com/archive/retrieve.cgi?id=SB96231664877464950
   5.djm
   Hyperlinks in this Article:
   (1) http://www.speechworks.com
   (2)
   http://interactive.wsj.com/archive/retrieve.cgi?id=SB96223832629236166
   7.djm
   (3)
   http://interactive.wsj.com/archive/retrieve.cgi?id=SB96214301754245574
   6.djm
   (4)
   http://interactive.wsj.com/archive/retrieve.cgi?id=SB96172311845639017
   6.djm
   (5)
   http://interactive.wsj.com/archive/retrieve.cgi?id=SB96145195951258990
   6.djm
   (6) http://www.quack.com
   (7) http://www.tellme.com/
   (8) mailto:[log in to unmask]
     _________________________________________________________________


VICUG-L is the Visually Impaired Computer User Group List.
To join or leave the list, send a message to
[log in to unmask]  In the body of the message, simply type
"subscribe vicug-l" or "unsubscribe vicug-l" without the quotations.
 VICUG-L is archived on the World Wide Web at
http://maelstrom.stjohns.edu/archives/vicug-l.html


ATOM RSS1 RSS2