BUSINESS ON THE WORLD WIDE WEB:

where "word of keystroke" begins

Nov. 1, 2001 -- Business uses of voice recognition


Transcript of the live chat session that took place Thursday, November 1, 2001. These sessions are normally scheduled for 12 noon-1 PM US Eastern Time (Standard Time = GMT -5, Daylight Savings Time = GMT -4) on Thursdays.

To connect to the chat room, go to www.samizdat.com/chat-intro.html

Since the chat itself happens at a rapid pace, it's often difficult to note interesting facts and URLs as they appear live. Here's a place to take a more leisurely look. I've rearranged some of the pieces to try to capture the various threads of discussion (which sometimes get lost in the rush of live chat).

Please send email with your follow-on questions and comments, and suggestions for topics we should focus on in future sessions. So long as the volume of email responses is manageable, I'll post the most pertinent ones here for all to see.

These sessions are hosted by Richard Seltzer. If you would like to receive email reminders of our chat sessions, simply send a blank email message to businessonthewebchats-subscribe@yahoogroups.com or go to http://groups.yahoo.com/group/businessonthewebchats and sign up there.

This is one of the longest-running chat programs on the Web. (Please let us know if you know of ones that are older.) We've been doing this since June 1996.

For transcripts of previous sessions and a list of future topics, www.samizdat.com/chat.html.

For an article on how to make "business chat" work (based on this experience), www.samizdat.com/events.html.

For articles on topics related to this one, check our newsletter, Internet-on-a-Disk www.samizdat.com/ioad.html


<

Participants

David
Ray Dee
Bill D. = Bill De Stefanis, vp of product development at Lernout*Hauspie (the company that bought Dragon) www.lhsl.com  (NB -- he used Dragon Naturally Speaking to input his comments in this chat session -- speaking instead of typing)
Doris
Bob Fleischer, Compaq, in New Hampshire
John Hibbs, Ben Franklin Institute of Global Learning, www.bfranklin.edu, San Diego, CA
TomHo
Richard Seltzer, host, www.samizdat.com, in Boston, Mass.
Bob Zwick, independent consultant in distance education, www.cottagemicro.com, in Texas

Threads of discussion (reconstructed afterwards)

Introductions
How Naturally Speaking works
What can you do with it?
Limitations -- Macintosh
Limitations -- multiple voices
Limitations -- input other than microphone
Telephone recognition and medical transcription
Company politics -- who develops what and who sells what?
Version 6 coming soon
Sony products
Factors affecting quality
Voice recognition and the blind
Future directions of voice recognition technology
Wrapup


Introductions

 11:58 - Richard Seltzer
       All, please introduce yourselves and let us know your interests.
       We'll be talking with Bill Destephanis from the makers of Dragon
       voice recognition products.

 11:59 - Richard Seltzer
       Hi, Doris. Where are you? What's your business? What's your
       interest in voice recognition?

 11:59 - Bob Fleischer
       I visited the exhibits at SpeechTEK 2001 in NYC last Friday -- L&H
       certainly had one of the more interesting booths. Of course, then I
       read a review that lists all of L&H's woes.

 12:00 - Bob Zwick
       I'm an independent consultant involved in Distance Education. I'm
       interested in finding out if voice to text software can be used to
       make a transcript of an Audio chat session.

 12:00 - Richard Seltzer
       Bill -- Please introduce yourself as well, and tell us about your
       role at Lernout*Hauspie, and tell us something about what you have
       available now and how people use your products and where you expect
       to take this to in the next stage. (That's a tall order. Just dive
       in with what feels comfortable as a starting point. Thanks.)

 12:00 - Bob Fleischer
       I'm particularly interested in speech on handheld devices (such as
       the iPAQ).

 12:01 - David
       This is David, and Im a consultant, but am primarily interested in
       speech as applied from a server perspective

 12:02 - doris     (Re: 11:59 - Richard Seltzer 'Hi, Doris. Where are you?
 What's your business? What's your...')
       I'm just a student who is interested in knew technologies. I'd like
       to know if the product is Mac compatible

 12:02 - Bill D.
       good afternoon Richard, to work in the product management
       organization here at Lernout & Hauspie, where I oversee
       development for all of speech and language products including the
       Dragon NaturallySpeaking product line.

12:18 - Hibbspc
       sorry to be late...breakfast meeting

12:20 - Richard Seltzer
       John -- welcome. I'd think that you'd be very interested in how to
       tie voice recognition into telephone-related apps.


How Naturally Speaking works


12:02 - Richard Seltzer
       Bob -- From the little I've used Dragon's Naturally Speaking 5, I
       suspect that this wouldn't be good for transcribing an audio chat
       session. You train the software to understand your voice. The more
       you train the better it gets. But it would be very difficult for
       the software to deal with multiple voices in normal conversation.

12:20 - Bob Zwick
       Richard - how much training time (hours) does it take to get accurate
       transcriptions ?

 12:21 - Richard Seltzer
       Bob -- I can't make an accurate estimate of time at this point.
       Need to test and train some more. Right now, I find it most useful
       in a mixed mode -- at least when composing. I can speak and do
       corrections/edits/change-of-minds manually as I go along, with
       keyboard and mouse.

12:20 - Bill D.
       Richard, in addition to training Dragon NaturallySpeaking (which
       for native speaker is less than 20 minutes) is also very helpful to
       allow the program to scan your written documents such as Microsoft
       Word or some of your sent to e-mail. By allowing this software to
       scan your existing documents it can extend its vocabulary and
       understand your writing style for how you combine words and phrases
       together.

 12:22 - Bob Zwick
       Bill D. - that is very usefull. Acuracy can get pretty good if it
       memorized my speach habits.

 12:22 - Richard Seltzer
       Bill -- that's an interesting twist -- training the system using
       your words, rather than someone else's words; because what matters
       isn't just your voice, but also your writing style and vocabulary.

 12:22 - Bill D.
       Richard, Dragon NaturallySpeaking comes in many versions or
       "flavors" as we call them :) that sell at different price points
       with different levels of functionality. The Essentials product you
       mention is our entry-level speech recognizer which is quite good
       for dictating into rich-text fields, like this chat session. E also
       sell higher and versions of the product, which allow it to be
       tightly integrated with office productivity applications such as MS
       office or Corel. This will allow users extended command and control
       be able to speech enable all of those productivity applications.

12:26 - Bill D.
       Richard, this notion of training the system both through training
       scripts and allowing a to read your documents is an excellent
       combination. Perhaps the most frequently overlooked aspect of
       training is allowing it to read old documents. Your writing style
       and my writing style do differ in the speech recognizer will make
       accommodations for that as well as extended its vocabulary for
       unique terms and phrases that you using your business. in version 6
       of Dragon NaturallySpeaking we made major changes to how the
       product improves accuracy by creating what we call an "Accuracy
       Center". This accuracy center, combines all aspects of training,
       microphone tuning, and vocabulary extension.

 12:28 - Richard Seltzer
       Bill -- what are the system requirements for all this sophisticated
       training? how much disk space is the data likely to require? I
       presume that the voice you speak in training is automatically
       deleted afterward and only the results are saved. But does that
       data take up significant space?

12:30 - Bill D.
       Richard, Dragon NaturallySpeaking, like most speech recognition
       products, is a fairly resource intensive application. The minimum
       system requirements for Dragon NaturallySpeaking are: a Pentium II
       processor, and 64 MB of RAM. Dispatch's requirements can dairy but
       a minimum of 110 MB is required in depending on the extent of your
       audiophile capture, this can rise to close to 300 MB.

 12:23 - Richard Seltzer
       All -- When you make corrections, you can also go into a training
       mode to teach the software particular expressions that you use
       often -- and how to spell them (e.g., the name of your company).
       The more often you do that, the better the results. And once
       trained, you really can speak at a normal pace -- which is faster
       than I can type.

 12:25 - Richard Seltzer
       All -- keep in mind that once you launch Dragon and turn on its
       microphone, your voice input becomes text output in a wide variety
       of applications -- whichever you current have as active. (It seems
       to work best with WordPad, but email and chat, etc. are also
       possible.)

12:16 - Richard Seltzer
       All -- when you are dictating into a document, there are a wide
       range of commands available, so you can speak not only text, but
       what to do with the text (like deleting and capitalizing and going
       back and selecting -- the full ranage; if you get into this, you
       don't need to type anything. You can even use voice to launch new
       applications, like Internet Explorer.

 12:17 - Richard Seltzer
       All -- while you can speak commands, personally, I find the
       application much more useful and efficient if you mix and match.
       You can type and speak and click -- using whatever method is
       simplest for what you are doing -- e.g., speak, then click and type
       to make corrections (instead of doing that by voice).

12:17 - Bill D.
       Richard, in addition to the commanding control you mention, Dragon
       NaturallySpeaking can also be scripted or programmed to perform
       automated tasks such as form-filling applications and working with
       structure documents.

       P.S. I am dictating this transcription with Dragon NatSpeak

12:31 - Richard Seltzer
       Bill -- are you speaking all of your input? or just occasionally?
       To give John Hibbs and others here a clearer sense of how good the
       accuracy can be (given that you are using a high-end version and
       have probably fine-tuned it to your voice and speak peculiarities).
 


What can you do with it?


 12:02 - TomHo
       I do interviews on video. I'd like a system I can use to transcribe
       the recordings. Do you have something that would be effective?

12:03 - Bill D.
       Tom, currently we do not have a technology that can transcribe a
       video recording as you mentioned. We do have a technology using
       speech recognition that can index and allow you to search for words
       and phrases against multiple videos or audio streams.

12:05 - Bob Zwick
       Richard - I suppose even me reading the text transcript would be
       faster than my typing it :-)

 12:07 - Richard Seltzer
       Bob Zwick -- That's an interesting possibility -- but if you already have
       a text transcript, why would you be using voice recognition to
       generate text?

 12:08 - Bob Zwick
       Richard - the text is only part of the audio chat. I suppose I
       could transcribe my audio and cut and paste the audience text
       portions.

 12:10 - Richard Seltzer
       Bob -- I think that if you try it, you will find interesting ways
       to take advantage of it, but not necessarily the kinds of things
       you are thinking of now. NB -- I got my 5.0 version (Essentials) for under $50
       from Broderbund. And it included a great high-quality microphone. Incredible price.
       A significantly better version (Preferred) now sells for $179.99.

12:12 - Richard Seltzer
       As an example of how well voice recognition can work and a very
       useful application, go to www.lhsl.com (their Web site), under
       About, go to their contact information. And give them a phone call.
       One of the first options is to go to their directory. There you
       simple state the name of the person you want to speak to and the
       system rings their line. Very slick. And the voice that speaks back
       at you, repeating the name that you spoke, sounds natural, not
       machine-like.

 12:09 - Richard Seltzer
       Bill -- I believe that it is difficult for those who haven't used
       voice recognition to understand both the limitations and the
       possibilities. Could you please describe some business applications
       that you know of?

12:13 - Bill D.
       Richard, most of the business applications for speech recognition
       have to do with large volumes of text creation what we call
       document creation services. Speech recognition's main benefit is
       productivity: it is faster to talk than to type.therefore you see
       higher volumes of speech recognition use in legal departments,
       government agencies, in various medical and General business
       practices where large amounts of document creation is done.

12:13 - Bob Zwick
       Richard - are you "speaking" the text we are seeing ?

 12:13 - Richard Seltzer
       Bob -- I was trying voice input to this area just before this
       session started and ran into a few problems -- but they were my
       problems, not problems with the software. I suspect that it would
       be best (if trying that) to click Pause before inputting, otherwise
       the automatic refresh gets in the way. I still have to play around and
       experiment some more to be able to speak-input here. Though it does
       seem to be possible. Bill -- are you speaking or typing now? If you
       are set up to speak, could you try that a bit so people get a sense
       of what the output would look like (with some random and inexact
       words every now and then).

 12:15 - Bob Zwick
       Richard - ahh.... automation getting in the way of automation :)

 12:17 - Bob Zwick
       Richard - do you see this as the new way for authors to write their
       manuscripts ?

 12:18 - Richard Seltzer
       Bob -- I'd like to experiment with this for writing -- fiction and
       articles. I need to do some more training though. The more you
       train the app the more accurate it is and hence the less time you
       have to spend making corrections. Think of this like scanning --
       you always need to proofread text that you've scanned -- only this
       is more like dealing with an early generation scanner when errors
       were more common.

  12:26 - Richard Seltzer
       I see this as a possible godsend for someone with carpal tunnel
       syndrome or arthritis, also for slow typers, also allowing
       handsfree use of the computer when you have to do many things at
       the same time.

12:28 - Richard Seltzer
       Bill, can you please tell us about other creative business uses of this technology?
       John -- Transcription is just one of many possibilities. You can
       also set this up to understand commands, etc.

 12:37 - Richard Seltzer
       Bill -- the more business applications you can tell us about the
       better. I'm sure there are important uses out there that I never
       dreamt of.

12:50 - Richard Seltzer
       Bill -- is there any place at the L&H Web site with demos or at
       least descriptions of a variety of business applications? Where
       should we go to learn more?

12:54 - Richard Seltzer
       Bill -- Is there a good place for people to see/experience demos of
       a variety of business apps or at least read about them?

13:00 - Bill D.
       Richard, yes there are several examples on our web site and I would
       also point you to the section called value added resellers (VAR.).
       Many of these folks, have done all kinds of interesting
       applications based on Dragon NaturallySpeaking.

 13:00 - Richard Seltzer
       Bill -- thanks. I'll have to take a close look at that VAR section.
       Great stuff. Thanks again.


Limitations -- Macintosh


12:03 - Richard Seltzer
       Bill -- following up on the question from Doris -- are your
       products PC-only? Or do they work with Macs as well?

 12:04 - Bill D.
       Doris, currently all of the Dragon applications are at its
       PC-based. We did not have products for the Macintosh platform.

 12:05 - doris     (Re: 12:04 - Bill D. 'Doris, currently all of the
 Dragon applications are at its...')
       Thank you, that effectively eliminates me


Limitations -- multiple voices


 12:03 - Bob Zwick
       Richard - I wonder just how bad multiple voices would be.

12:04 - Richard Seltzer
       Bob Zwick -- I think you'd have to experiment to know for sure. But I
       wouldn't have very high hopes. On the other hand, this could be
       very good for transcribing any voice that is a single voice, if the
       speaker had previously "trained" the software.

 12:03 - Bob Fleischer
       I believe L&H has products that are speaker independent and
       intended for converting large volumes of audio to text, right?

12:06 - Bill D.
       Bob F., Lernout & Hauspie does not currently sell a product that is
       speaker independant which would allow you to transcribe large
       volumes of audio and text. In our research labs we are working on
       just this problem and it's a derivative of a large vocabulary
       speaker engine like Dragon.

12:07 - TomHo     (Re: 12:03 - Bill D. 'Tom, currently we do not have a
 technology that can...')
       Bill, What is the name of the software you mentioned?

12:08 - Ray Dee     (Re: 12:07 - TomHo 'Bill, What is the name of the
 software you mentioned?...')
       Tom Ho -- Do you know of a similar system?

12:08 - Bill D.
       Tom H. The product used to indexed Digital media is called
       MediaIndexer we shipped our first version of this product in July
       of 2001.

2:25 - Hibbspc
       I was thinking that Dragon was more like the text you see on t.v.
       where the voices are transcribed "on the fly" for the hearing
       impaired. Is that part of what your company offers?

 12:28 - Hibbspc
       I was hopeful to see "it" this way - during events, have the
       software "listen" to the speakers for automatic, instant text
       traslations which were (either or both) uploaded in the original
       language and/or machine translated into other languages. No?

 12:28 - Bob Zwick
       BillD. - does Rush Limbaugh use you product to see what his radio
       callin people are saying ?

 12:28 - Bill D.
       Hibbspc, No, the Teleprompter unit to see on television is
       typically done manually by human beings. Naturally speech
       recognition is being looked at to augment this process and allow it
       to fully transcribed feature like broadcasts.Currently, there are
       some research challenges we have to overcome, such as music
       overlays with spoken text: which really confuse the speech
       recognizers today.

 12:31 - Hibbspc
       Is the bottom line, at least for now that each individual speaker
       has to "train" the software, otherwise the text is hopelessly
       inaccurate?

 12:33 - Richard Seltzer
       John Hibbs -- Yes, I'd say that the training is absolutely
       essential, at least at today's level of development. What's your
       take on that, Bill?

12:34 - Bill D.
       Hibbspc, large vocabulary recognition products, like Dragon
       NaturallySpeaking, do require training. The notion of speaker
       independent recognition, is probably several years away. We work
       extensively here in our research department on just this problem
       and have reasonable accuracy for certain applications but we are
       not prepared to make this commercially available today.

 12:39 - Hibbspc
       As I understand it, we could not take our audio archives and run
       thrm through Dragon and expect to have a reasonably accurate
       transcription of the dialogue? n our case, the audio was capture d over the telephone and
       "delivered" directly to the p.c. where it (remains) stored. That
       audio would not be transcribealbe - not that I am sure anyone wants
       it!
 12:41 - Richard Seltzer
       John -- I think that the main consideration would be whether there
       was one voice or multiple voices. If one voice, then it's possible.
       But the more different voices, the lower the accuracy, quickly
       deteriorating to useless. Is that accurate, Bill?

  12:40 - Bill D.
       Hibbspc, yes, search for a product called PowerScribe.

 12:41 - Hibbspc
       thanks re jpowerscribe

12:34 - Bob Zwick
       What about the high end product mentioned earlier. The
       one that is used for mass transcription?
       Bill can you tell us a little more about that product?

12:42 - Bill D.
       Richard, multi-voice transcription is still a ways off (probably
       two to three years) what we can do today, is index a multi-voice
       audio file. With the product I mentioned earlier, MediaIndexer

 12:43 - Richard Seltzer
       Bill -- Are you working with any of the Web search sites or
       producers of search software to make it possible to index voice
       content and therefore make it searchable? e.g., are you working
       with AltaVista?

 12:46 - Bill D.
       Richard, yes, we have spoken with various Web content sites who
       have digital audio content that would like it to be indexed and
       made searchable. we also talking with various General business, and
       universities, who may have training video content.

 12:47 - Richard Seltzer
       Bill -- You might also want to check with the folks at Vanderbilt
       U. who have an archive of all the network news shows on television
       from back in the days of the Viet Nam War. Indexing the voice of an
       archive like that could be very useful. Also indexing the many
       audio files stored by the Library of Congress.


Limitations -- input other than microphone


 12:05 - Richard Seltzer
       Bill -- if the individual who recorded the voice also had
       previously "trained" the software to his/her voice, wouldn't that
       kind of transcription be possible?

12:29 - Hibbspc
       Where can I see this work in a real live demonstration. You
       mentioned Rush Limbaugh - does he use it some way?

 12:29 - Richard Seltzer
       Bob -- I strongly doubt that Rush could use this that way. It's the
       same problem -- a random caller wouldn't have had a chance to train
       the system, hence the accuracy of the recognition would be
       frustratingly low. Is that right, Bill?

 12:30 - Hibbspc
       re: 12.28 thanks..that helps. Bill was H&L working on that before
       the big money problems?

12::32 - Bill D.
       Bob Z. I am not aware that Rush Limbaugh uses Dragon
       NaturallySpeaking. however given his current condition it may be of
       some use.However, telephone recognition is not merely as accurate
       as that through direct microphone imput.

 12:40 - Bob Zwick
       Does the input to DragonSpeak have to be a mic ? Can input be
       redirected from software ie. Real Audio, or audio file players.

 12:42 - Richard Seltzer
       Bill -- yes, I'm curious too. Is there a way (with your consumer
       products) of feeding a different audio source (other than a mic)
       into the voice recognition software.

 12:44 - Bill D.
       Richard, in addition to correct microphone transcription (like I'm
       doing right now) Dragon NaturallySpeaking can transcribe recorded
       audio file such as WAV files. these WAV. files will have to be
       recorded by someone who is trained on the system i.e. someone that
       went through Dragon training.That would be the extent of its
       abilities today.

12:46 - Richard Seltzer
       Bill -- Does it have to be WAV files? Those can easily be enormous
       -- aobut ten times the size of MP3. Would MP3 files work?
 12:49 - Bill D.
       Richard, yes even though they are large, WAV files are currently
       the only format that we support: no MP3.

 12:45 - Richard Seltzer
       Bill -- as an interim measure, if a limited number of people are
       typically involved in voice conversations that need to be
       transcribed, is it possible for each member of that small group to
       train the software to their voices and styles, and then for the
       system to recognize the speaker and adjust automatically, or for
       the speakers to start what they say with their name as a way of
       shifting the software to their profile? Do you have anything like
       that now? or are you working on it?

12:46 - Bob Zwick
       Richard - that would be an interesting experiment. I could train
       your software for my voice from an audio chat room.

12:48 - Richard Seltzer
       Bob -- yes, that would be an interesting experiment. But we'd want
       the software to be able to switch from looking for one speaker
       profile to another as speakers changed in the course of a
       conversation.

 12:47 - Bill D.
       Richard, just clarify the abilities of Dragon NaturallySpeaking; it
       currently cannot do multi-voice transcriptions, even if they are
       trained voices. We do not have the ability to extract separate
       voice profiles from 1 audio stream.

 12:49 - Bob Zwick
       Bill D. - I understand that (no multi-voice), but can a system have
       multiple personallities that can be activated ?

 12:49 - Richard Seltzer
       Bill -- Understood that what I'm speculating about now isn't
       current possible. But shouldn't it be possible to add a voice
       command to your command file that would change speaker profiles? So
       a new speaker could give that command and hence be understood?
       Feels like a short-term development project that could be useful.

  12:50 - Bill D.
       Bob, yes Dragon NaturallySpeaking can support many voice profiles
       with one piece of software. Typically it is licensed to one
       individual who may have several voice files recorded for different
       types of microphones, environments where they may record, or for
       mobile recording.

12:51 - Richard Seltzer
       Bill, I think that what Bob was looking for (me too), was a way to
       shift from one profile to another, on the fly.

 12:52 - Bob Zwick
       Bill D. - that's great. Can mutiple instances of the program be
       running on a system too?

12:32 - Richard Seltzer
       Bill -- how long have you been working in the voice recognition
       field? I remember Solzenitsyn's First Circle and what it had to say
       about voice prints. Also, Gordon Bell, who was Digital's computer
       guru back in the 1970s and early 80s began his computing career
       focusing on voice recognition, but got out of it when he realized
       how very very difficult it was to do it well.

 12:52 - Richard Seltzer
       Bill -- Thinking again of First Circle by Solzhenitsy, does the
       training with Dragon generate some kind of a voice print? Could
       well-trained dragon software be used determine that a speaker was
       in fact who he/she said he/she was?

 12:54 - Bill D.
       Richard, the "voiceprint" you mention is something we refer to in
       the industry as speaker verification: an example of biometric
       technology. This is separate and distinct from Dragon
       NaturallySpeaking and we sell technologies in the speaker
       verification space to be used for identifying users for security
       purposes..

 12:54 - Bob Fleischer
       There are a number of vendors who sell the terchnology and service
       for voice authentication -- DNS may not be the tool for it, but it
       does exist.

12:56 - Richard Seltzer
       Bill -- I'd think that that speaker verification software could be
       combined with your Naturally Speaking to recognize when speakers
       change in multi-speaker situations -- but that's probably years in
       the future...

12:57 - Richard Seltzer
       Bob Fleischer -- in a short while we'll be identifiable in so many
       ways it's ridiculous (DNA too). I bet it would be possible to
       profile someone based on their typing style and the vocabulary they
       use in typing, so you could know that the person who is in the chat
       room is in fact the person he/she says...


Telephone recognition and medical transcription


2:34 - Hibbspc
       ah, the "telephone recognition" area was where my talks centered
       about 24 months ago..I guess that has not moved forward very
       much...understandable as I would think very low demand for same.

 12:37 - Bill D.
       Hibbspc: actually, telephone recognition has made great strides the
       last several years. We in fact have developed a product
       specifically for medical transcription. It is sold through a
       special distribution channel to the medical profession. It allows
       for telephone based transcription of medical analysis.the problem
       of telephone based recognition becomes very manageable if you can
       limit what is called the domain i.e if I know what you are going to
       talk about my degree of accuracy will improve dramatically.

12:38 - Hibbspc
       Bill, that product you talked about for the medical industry..is
       that on your web site?

 12:38 - Richard Seltzer
       Bill -- in that medical transcription application, do you mean that
       if you limit the vocabulary -- the allowed words -- then the need
       to train to a particular voice goes down considerably?

 12:39 - Bill D.
       Richard, mobile transcription, for instance capturing recordings on
       digital audio is best done with a high-quality machine to begin
       with. There is no need to send the audio file through a microphone
       for transcription if it's captured at first high-quality manner it
       can be transcribed directly against the speech recognizer without
       going through microphone imput.

12:40 - Richard Seltzer
       Bill -- aside from the medical, do you have other examples of
       multi-voice but limited vocabulary applications? (perhaps the voice
       recognition built into the phone system at L&H is an example like
       that).


Company politics -- who develops what and who sells what?


 12:06 - Richard Seltzer
       Bill -- I understand that your consumer products, like the one I
       purchased, are now sold by Broderbund. Do you still handle the
       development of the whole Dragon line -- including consumer?

  12:07 - Bob Zwick
      Can you point us to where we can find out more about L&H ?

 12:07 - Bill D.
       Richard, all development of Dragon products are done here at
       Lernout & Hauspie; the Broderbund relationship you mentioned is
       purely a retail distributor relationship.

  12:08 - Bob Fleischer
       www.lhsl.com is mentioned in the literature I got at SpeechTEK

12:20 - Hibbspc
       What was the impact of L&H going under due to the problems in
       Europe with the buy out? are you now completely independent?
       operating out of bankruptcy?


Version 6 coming soon


12:08 - Richard Seltzer
       Bill -- I understand that you just came out with a new version of
       Naturally Speaking -- 6.0. How does that differ from 5.0? Is it
       just a matter of improved quality of voice recognition? Or are
       there functions you can perform with it that you couldn't with 5.0?

12:11 - Bill D.
       Richard, the new version of Dragon NaturallySpeaking, version
       6.will be available later this month. As with all new versions of
       speech recognition it does have significant accuracy improvements
       and new features such as "Nothing but Speech" which allows speech
       recognizer to ignore extraneous sounds like "ahhs" and "Umms" that
       we all do what I natural conversation.

  12:19 - Richard Seltzer
       Bill -- Excellent. How does Dragon NatSpeak differ from the
       bare-bones, consumer software (Naturally Speaking 5.0, Essentials)
       that I'm using?


Sony products


 12:11 - Bob Fleischer
       I also noticed at the show that Sony was pitching a line of digital
       memo recorders that could produce input directly to DNS.

 12:15 - Bill D.
       Bob F. , Lernout & Hauspie has a development relationship with Sony
       where we have done extensive work on speech recognition integration
       with their digital audio recorders. Dragon NaturallySpeaking can
       transcribed any high-quality WAV. file, assuming the user has been
       trained for the system.



 

Factors affecting quality


12:34 - Richard Seltzer
       Bob -- yes, the microphone matters a lot too. You need a good one
       (which comes with the software), and you need to place it right and
       consistently (with a headset -- the mouth being a consistent
       distance from your mouth). Other pointers, Bill?

  12:35 - Richard Seltzer
       Bill -- given the importance of the microphone and positioning of
       the microphone, would it make sense for an author to record/dictate
       onto cassette (virtually anywhere) and then play the cassette into
       a mic/PC running Dragon in order to get a rough transcript?


Voice recognition and the blind


 12:34 - Hibbspc
       My 84 year old mother, now nearly blind and good typist, but no
       longer...is she your market?

 12:37 - Richard Seltzer
       Bill -- do the blind or people whose sight is failing use this
       product? I suspect it would be difficult for them, because of the
       need to read the screen to check for errors and correct them. On
       the other hand, it would be great for anyone who has trouble typing
       -- either just plain slow or due to arthritis, etc. I'm thinking of
       giving this to my mother (81) who is having increasing difficulty
       dealing with a keyboard.


Future directions for voice recognition technology


 12:50 - Richard Seltzer
       Bill -- What do you see as the most important/useful directions for
       future development (aside from dealing with multiple voices)?

 12:52 - Bill D.
       Richard, the biggest research tasks we have with speech
       recognition, is to make it far more natural. Several things we are
       looking at along these lines include: automatic punctuation,
       speaker independence (no training), and of course multi-voice
       conversation as you mentioned.

 12:53 - Richard Seltzer
       Bill -- I'm really amazed at what is possible today -- and at an
       affordable price. I had thought that this technology was much
       farther from prime time. I had thought that only applications with
       severely limited vocabularies made sense today. I'm looking forward
       to experimenting more with my version 5.0.
 



 12:54 - Bob Zwick
       What applications will DragonSpeak NOT work with ? ie. AOL, Java,
       MS Word 2000 etc.

12:56 - Bill D.
       Bob, Dragon NaturallySpeaking will work with virtually any Windows
       based application. Any application that allows for standard text
       imput can be speech enabled with Dragon NaturallySpeaking. We
       specifically add commanding control functions, for limited set of
       office productivity applications, Web tools, and various e-mail
       programs.

12:58 - Bob Zwick
       Bill D. - that's interesting. I could be speaking in one audio chat
       room and have it transcribed into another text chat room !
 

 12:58 - Richard Seltzer
       Bob Zwick and others -- the command capabilities of this software
       seem to make it natural for embedding it, combining it with other
       Windows apps in creative ways. I'd love to learn more about what
       has already been done along that line. Anything like that at your
       Web site, Bill?

 12:59 - Richard Seltzer
       Bob Zwick -- I could imagine that combination of audio and voice
       chat being very useful -- even with PalTalk -- to talk and have
       what you say automatically entered in the text chat window. Would
       be great for Global Learn Day next year so folks without voice
       output could still get a pretty good sense of what's going on.


Wrapup


 12:55 - Richard Seltzer
       All -- we're just about at the end of the hour. Please, before
       signing off, enter your email and URL addresses so we can keep in
       touch. And please join us again next Thursday. Bill and all, thanks
       very much for joining us today.

  12:57 - Bob Zwick
       Bob Zwick http://www.cottagemicro.com/coinfo/contact.htm
       Voice/Text contact

  12:57 - Bill D.
       Richard, it has been a pleasure, thanks so much for inviting me.
       This is the first chat session where a been so actively involved.
       It's been fun! My e-mail address is BDESTEFANIS@LHSL.COM

  13:00 - Richard Seltzer
       Bill, thanks again. All -- time to wrap up. Yes, do please enter
       your contact info here before signing off.

 13:01 - Bill D.
       Richard, thank you, when a transcription of this conversation is
       posted please let me know thanks again Bill DeStefanis

 13:02 - Richard Seltzer
       Bill -- Will do. Thanks again.

 13:02 - Bob Zwick
       Richard - be sure to listen to my interview on TheReport.Com AM
       Radio on Nov. 13 6:30 PM EST

 13:03 - Richard Seltzer
       Bob Zwick -- Glad to hear that you hooked up with Bill and Dave.
       Yes, I will try to tune in.

 13:05 - Bob Zwick
       I really like the multiple personalities. I can ahve one for a
       phone bridge, a mic, an audio chat room and mp3 files !

 13:06 - Bob Zwick
       Thanks Bill

 13:06 - Richard Seltzer
       Bob -- Interesting idea. Yes, that should improve the accuracy of
       the transcriptions. You ought to give it a try.

 13:06 - Bob Zwick
       bye all

 13:06 - Richard Seltzer
       Bye.



Previous transcripts and schedule of upcoming chats -- www.samizdat.com/chat.html

To connect to the chat room, go to www.samizdat.com/chat-intro.html

The full text of Richard Seltzer's books The Social Web, Take Charge of Your Web Site, Shop Online the Lazy Way, and The Way of the Web, plus more than a hundred related articles are available on CD ROM My Internet: a Personal View of Internet Business Opportunities.

Web Business Boot Camp: Hands-on Internet lessons for manager, entrepreneurs, and professionals by Richard Seltzer (Wiley, 2002). No-nonsense guide targets activities that anyone can perform to achieve online business
success. Reviews.

a library for the price of a book.

This site is Published by B&R Samizdat Express, 33 Gould St., West Roxbury, MA 02132. (617) 469-2269. seltzer@samizdat.com


Return to B&R Samizdat Express


<


Internet Business Showcase: