OS/2 Shareware BBS: 14 Text/14-Text.zip/arc_7e.zip/SPCHOUT.SAV

Text File | 1994-04-16 | 120KB | 3,658 lines

This file uploaded by John Pedersen 76547,357.

Messages collected using Golden Commpass 2.1

Messages sorted with Recon 2.2



#: 1 S10/Thread Header
    08-Feb-94  16:10:46
Sb: Speech quick interaction
Fm:
To:

TITLE: SPEECH QUICK INTERACTION
KEYWORDS: QUICK RESPONSE SPEED PROFILE

#: 705 S10/IBM Speech Products ∙5 ∙
    08-Feb-94  16:10:46
Sb: Speech quick interaction
Fm: Dave Briccetti [Zif 72241,557
To: All

Hello.

I'm using ICSS for the first time, and I like it very much so far.  I'm
looking for pointers to information that will allow me to tune my ICSS
applications for very quick response.  For simple responses, such as "Yes" and
"No," I'd like ICSS to return to my application within 1/2 second after the
speaker stops talking.  Is this done entirely by adjusting the system
parameters, or is there something else I need to do?.

Thanks, and keep up the good work.

Dave Briccetti
Dave Briccetti & Assoc.
OS/2 Consultant/Developer
Lafayette, CA


#: 725 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  16:39:51
Sb: #705-Speech quick interaction
Fm: Ora Williamson 71742,2161
To: Dave Briccetti [Zif 72241,557

>> For simple responses, such as "Yes" and "No," I'd like ICSS to return to my
application within 1/2 second after the speaker stops talking.

Dave: In the User Profile, adjust the Minimum Amount of Silence To Indicate
End Talking variable.  Or, adjust it with the ICSSSetValue API call using the
UttEndSilDuration variable.

Hope this helps! Regards, Ora J. Williamson IBM Spoken Language Systems Lab
ICSS Technical Liason

#: 1 S10/Thread Header
    27-Jan-94  23:56:29
Sb: Speech Products
Fm:
To:

TITLE: SPEECH PRODUCTS
KEYWORDS:  ICSS IPDS

#: 575 S10/IBM Speech Products ∙5 ∙
    27-Jan-94  23:56:29
Sb: Speech Products
Fm: R. Glenn Jones 73531,461
To: Ian Wright 73064,3434

Ian, I am a little confused about IBM's various speech products. Was the IBM
Dictation system written using ICSS? I have purchased the dicatation system
(have the software but not the card yet) and wonder now if I have bought the
right product.

Can you tell me what the differences are between the various products?

Thanks R. Glenn Jones (Sears Canada Inc)


#: 580 S10/IBM Speech Products ∙5 ∙
    29-Jan-94  11:46:38
Sb: #575-Speech Products
Fm: Ian Wright (IBM) 73064,3434
To: R. Glenn Jones 73531,461

>>I am a little confused about IBM's various speech products

Well, me too.  But here's what I have found.

IBM Personal Dictation System ------------------------- * Discrete speech
dictation * 20,000-word vocabulary (World News Today reported 32,000-word
vocab, 20 Nov 93) * 70 words a minute * Needs training * Continuous speech
commands (Solitaire's great fun) * Voice Action Editor to create personal
macros (e.g. to open e-mail when you say "Good morning") * User interface
navigation

IBM Speech Server Series -------------------- * Similar to IPDS in function *
Runs under AIX (host and client) or AIX host and OS/2 client * High-end
solution includes networking

IBM Continuous Speech Series ------------------------ * Speaker-independent *
Continuous speech * Toolkit for developing Windows, OS/2, and AIX
speech-enabled apps


VoiceType 2 ---------- * Supports DOS apps such as word processord, databases,
and spreadsheets

Voice Type Control for Windows -------------------------- * 64-word vocabulary
command and navigation system for Windows 3.1

For Information ------------ Call 1-800-TALK-2ME Ian Wright (IBM)


#: 598 S10/IBM Speech Products ∙5 ∙
    01-Feb-94  14:00:27
Sb: #575-Speech Products
Fm: Ora Williamson 71742,2161
To: R. Glenn Jones 73531,461

>>I am a little confused about IBM's various speech products

Time to chime in...

IBM's 2 PRIMARY Speech products are IBM Personal Dictation System (IPDS) and
IBM Continuous Speech Series (ICSS).

IPDS is a large-vocabulary dictation system which allows you to dictate after
performing an 1 1/2 hour enrollment where you train your voice to the system.
IPDS supports discrete speech - there must be a slight pause between words.
IPDS is currently available for OS/2 and runs about $1000 for the software and
the speech accelerator card.

ICSS is a small-vocabulary grammar driven system which allows you to speak
pre-defined words and phrases continuously - without pauses between your
words.  ICSS is speaker-independent - you don't have to train your voice to
the system.  It is great for walk-up-and-use apps.  ICSS ships with an SDK for
speech-enabling applications.  The DLL supplies 12 'C' function calls. For
each application, you can load up to 128 of these grammars (1000 words max per
grammar) and dynamically switch between them, typically from window to window.
ICSS is currently available on OS/2, AIX and Windows.  It supports sound cards
such as the SoundBlaster, Pro Audio Spectrum and IBM Audio Capture and
Playback Adaptor.  ICSS costs $315 for a single development license (which
comes with a runtime license).

For information on ANY of IBM's Speech products, call 1-800-TALK2ME.  To order
ICSS, call Indelible Blue at 1-800-776-8284.

Ora Williamson ICSS Technical Liason IBM Spoken Language Systems Lab


#: 604 S10/IBM Speech Products ∙5 ∙
    01-Feb-94  16:47:42
Sb: #598-Speech Products
Fm: Scott Gurvey 70040,156
To: Ora Williamson 71742,2161

Ora...

Thanks for making the distinction between the speech products clear. Now I
think I've got it. I sure confused them in the first story I did last year. Do
you know if the Personal Dictation Adaptor can serve as the audio input device
for ICSS? SO far the only audio board I have for my PS/2 95 is the
Audiovation/A, and there is no OS/2 support for that.

Cheers,
 Scott


#: 611 S10/IBM Speech Products ∙5 ∙
    01-Feb-94  23:37:06
Sb: #604-Speech Products
Fm: Brandon Booth [IBM] 71042,3144
To: Scott Gurvey 70040,156

ICSS can use ACPA, Sound Blaster, and Pro Audio Spectrum cards for audio
input.

Brandon

ICSS Development




#: 622 S10/IBM Speech Products ∙5 ∙
    02-Feb-94  13:23:16
Sb: #611-Speech Products
Fm: Scott Gurvey 70040,156
To: Brandon Booth [IBM] 71042,3144

>ICSS can use ACPA, Sound Blaster, and Pro Audio Spectrum cards for audio
input.

So.... Can the IPDS card emulate any of those? Can the Audiovation or
Windsurfer? Will ICSS work with either of the last two (the MWave cards) once
OS/2 support is available?

Does the idea of needing three IBM audio boards on a computer to support three
different IBM products strike anyone else as a little silly?

Cheers,
 Scott

#: 677 S10/IBM Speech Products ∙5 ∙
    06-Feb-94  20:52:35
Sb: #622-Speech Products
Fm: Brandon Booth [IBM] 71042,3144
To: Scott Gurvey 70040,156

Scott

We should be able to use any card supported by MMPM, and so if Windsurfer is
so supported, we should be able to use it.  I really can't comment on our
relationship to IPDS except to say that the two products really do solve
different problems (and is perhaps not as optimal as one would like).

Brandon

ICSS Development



#: 700 S10/IBM Speech Products ∙5 ∙
    08-Feb-94  14:35:56
Sb: #598-Speech Products
Fm: R. Glenn Jones 73531,461
To: Ora Williamson 71742,2161

Ora, The exchange rate must be a little higher than I thought. IBM Canada is
charging roughly $800 each for the software and adapter. I bought IPDS because
I intend to use it as a means of entering data and to a lesser extent to
control programs. Voice control for Windows was mentioned, is this written in
ICSS? R. Glenn Jones (Sears Canada Inc)

#: 716 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  10:15:31
Sb: #700-Speech Products
Fm: Ora Williamson 71742,2161
To: R. Glenn Jones 73531,461

>> IBM Canada is charging roughly $800 each for the software and adapter.

Glenn: The $800 is for the cost of IPDS, not ICSS, right?  IPDS was not part
of the $79 special - only ICSS.

>> Voice control for Windows was mentioned, is this written in ICSS?

I don't understand the above question.  Are you asking about ICSS for Windows,
or ICSS for OS/2?

Regards, Ora J. Williamson IBM Spoken Language Systems Lab ICSS Technical
Liason

#: 738 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  23:01:14
Sb: #716-Speech Products
Fm: R. Glenn Jones 73531,461
To: Ora Williamson 71742,2161

Ora, yes those figures were for IPDS.

I was asking how Voice Control for Windows fit into the picture. I was asking
if Voice Control for Windows was an application that IBM has written using
ICSS. R. Glenn Jones (Sears Canada Inc)

#: 1 S10/Thread Header
    06-Feb-94  22:07:02
Sb: ICSS WIN Problem
Fm:
To:

TITLE: ICSS WIN PROBLEM
KEYWORDS: ICSS WINDOWS TEST BANDWIDTH

#: 679 S10/IBM Speech Products ∙5 ∙
    06-Feb-94  22:07:02
Sb: ICSS WIN Problem
Fm: Dan Devine 76040,215
To: all

I have just installed ICSS for Windows.  The installation went smoothly on my
Gateway 486-66 w/ 16M.
I'm getting two errors from what is probably the same problem.  When I try to
run the test I get the
following error when I ACTIVATE the TESTING.CTX file:

LoadContext failed, rc:49, Recognition processing could not open the
associated CPDFILE.

The error appears when I try to activate the other ctx files as well.  I also
get the second half of this
error (Recognition..CPDFILE) when I load the Stock Market demo.

I have a PAS16 card that functions properly.  Any help would be appreciated.


DAN

#: 689 S10/IBM Speech Products ∙5 ∙
    07-Feb-94  16:41:34
Sb: #679-ICSS WIN Problem
Fm: Ora Williamson 71742,2161
To: Dan Devine 76040,215

>> LoadContext failed, rc:49, Recognition processing could not open the
associated CPDFILE.

Dan:

Do me a favor and look for the file MEDIUM.CPD in the following directory:
\ICSSWIN\MODEL\US_ENG\MEDIUM.

If it is in there, then you may have a path problem in the AUTOEXEC, or your
Virtual Memory size is not large enough.  Set Virtual Memory to Permanent and
around 12 - 15 Meg.

If you don't have the .CPD file, then I will send it to you.

Ora Williamson IBM Spoken Language Systems


#: 742 S10/IBM Speech Products ∙5 ∙
    10-Feb-94  11:28:19
Sb: #689-ICSS WIN Problem
Fm: Dan Devine 76040,215
To: Ora Williamson 71742,2161

Ora,

Thanks for your help.  The problem resulted because I had installed the
high banwidth models only, not the medium -  which the included contexts had
been created with.

Thanks again,

Dan

#: 814 S10/IBM Speech Products ∙5 ∙
    16-Feb-94  13:01:28
Sb: #742-ICSS WIN Problem
Fm: Ora Williamson 71742,2161
To: Dan Devine 76040,215

>>The problem resulted because I had installed the high banwidth models only,
not the medium

Dan: We are updating our documentation to deal with this anomoly.  Thanks for
your patience!

Ora.

#: 1 S10/Thread Header
    17-Feb-94  09:31:22
Sb: ICSS Prob Determination
Fm:
To:

TITLE: ICSS PROB DETERMINATION
KEYWORDS: LIST PROBLEMS SOLUTIONS

#: 832 S10/IBM Speech Products ∙5 ∙
    17-Feb-94  09:31:22
Sb: ICSS Prob Determination
Fm: Ora Williamson 71742,2161
To: all

To ICSS Users:

I have uploaded a file in our section ICSSPD.1 for your perusal.  This file is
a list of common problems that we have seen that you may be encountering.
Possible solutions are included.  As the list grows, we will upload additional
hints/tips in sequential file order (ICSSPD.2, ...).

Hope this is helpfulRegards, Ora J. Williamson IBM Spoken Language Systems Lab
ICSS Technical Liasion

#: 1 S10/Thread Header
    17-Feb-94  19:21:52
Sb: ICSS API for Phonemes
Fm:
To:

TITLE: ICSS API FOR PHONEMES
KEYWORDS: RETURN PHONEMES

#: 846 S10/IBM Speech Products ∙5 ∙
    17-Feb-94  19:21:52
Sb: ICSS API for Phonemes
Fm: Tom Vreeland 76711,1075
To: Brandon Booth [IBM] 71042,3144

We are currently extending our MEDIAscript OS/2 myltimedia authoring tools to
support speech recognition with ICSS. This will enable developers to use our
tools to build Speech enabled multimedia applications without any C
programming.

One important API which is missing from the current support in V 1.0 is the
ability to get a phoneme string back in addition to the text string. This is a
key element in the strategies we would like to use to extend the capabilities
of ICSS. Does such an API exist internally? Could it be provided?

This would facilitate lexicon and dictionary construction, and support many
uses in linguistic research using this technology.

NOTE: We have also added pen support to our authoring interface and like the
idea that we can load a gesture set and sr context for a given application
element to support both elements of an enhanced mm interface.

#: 849 S10/IBM Speech Products ∙5 ∙
    17-Feb-94  22:17:44
Sb: #846-ICSS API for Phonemes
Fm: Bob Cave 73232,2474
To: Tom Vreeland 76711,1075

Tom,

As far as I know, we have no internal API to get the phoneme string back along
with the recognized text.  I will double check this and get back to you (may
not be until next week as I will be out of the office tomorrow).

Regards,

Bob Cave, ICSS Development Team

#: 892 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:23:41
Sb: #846-ICSS API for Phonemes
Fm: Brandon Booth [IBM] 71042,3144
To: Tom Vreeland 76711,1075

Hi Tom

The stuff you're describing with the multimedia authoring sounds cool.  I'm
sorry we don't have any api to return the phoneme string with the reco'd text.
 You could read our dictionary (icss_us.dic) to get this information.  There
are no plans to add such an api to ICSS.

Brandon

ICSS Development - IBM

#: 1 S10/Thread Header
    08-Feb-94  14:55:06
Sb: Other Language Lexicons
Fm:
To:

TITLE: OTHER LANGUAGE LEXICONS
KEYWORDS: LANGUAGES FOREIGN SPEAKER ACCENTS RECOGNITION

#: 702 S10/IBM Speech Products ∙5 ∙
    08-Feb-94  14:55:06
Sb: Other Language Lexicons
Fm: James E. Osbourn 74150,2571
To: Brandon Booth 71042,3144

Are there ICSS dictionaries/lexicons for other languages than English?  If
yes, which ones? are they available in the US?


James in Kansas City

#: 726 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  16:39:59
Sb: #702-Other Language Lexicons
Fm: Ora Williamson 71742,2161
To: James E. Osbourn 74150,2571

>>Are there ICSS dictionaries/lexicons for other languages than English?

James: No, there are not, but you can emulate foreign phrases using the
English phoneme set.  Add the foreign words to an addendum dictionary and
create the associated phoneme strings with the U.S. English phonemes provided
with ICSS.
 When doing this, it is best to create the phrases as 1 long word.  For
example:

MUCHAS_GRACIAS                    M UW  CH AA S  G R AX  S IY AA S

Hope this helps! Regards, Ora J. Williamson IBM Spoken Language Systems Lab
ICSS Technical Liason

#: 733 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  21:13:57
Sb: #702-Other Language Lexicons
Fm: Brandon Booth [IBM] 71042,3144
To: James E. Osbourn 74150,2571

James

I'm sorry, we currently only support American English.

Brandon

ICSS Development



#: 748 S10/IBM Speech Products ∙5 ∙
    10-Feb-94  12:27:02
Sb: #733-Other Language Lexicons
Fm: Bob Hutchison 72254,1315
To: Brandon Booth [IBM] 71042,3144

Brandon,

Can you clarify this a bit.  What part of America?  The US has such a wide
range of accents that I cannot quite imagine what you mean.  Certainly in
written english there are differences between American English and the rest of
us, but oral?

Bob


#: 829 S10/IBM Speech Products ∙5 ∙
    16-Feb-94  20:03:18
Sb: #748-Other Language Lexicons
Fm: Ken Solkowitz 72103,2111
To: Bob Hutchison 72254,1315

Bob,
With ICSS' speaker independent aspects, it recognizes almost anyone in   the
U.S. I've demonstrated at Comdex for the past two years and it recognized
everyone fine. It truly is speaker independent.

Regards,
Ken Solkowitz
IBM Continuous Speech Series
Bethesda, MD
301-803-3897

#: 833 S10/IBM Speech Products ∙5 ∙
    17-Feb-94  10:36:58
Sb: #829-Other Language Lexicons
Fm: Bob Hutchison 72254,1315
To: Ken Solkowitz 72103,2111

Ken,

That's what I thought.  Thanks for confirming.

Bob

#: 872 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  13:09:34
Sb: #829-Other Language Lexicons
Fm: Michael Burgun 100020,1604
To: Ken Solkowitz 72103,2111 (X)

Ken,

<<<Bob, With ICSS' speaker independent aspects, it recognizes almost anyone in
 the U.S. I've demonstrated at Comdex for the past two years and it recognized
everyone fine. It truly is speaker independent. >>>

What about people without funny accents like us Australians?!!?

Michael

#: 874 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  15:13:52
Sb: #872-Other Language Lexicons
Fm: Ken Solkowitz 72103,2111
To: Michael Burgun 100020,1604

Michael,
Actually we have a harder time with UK English speakers than we do with anyone
else. We are working on this by recording UK English speakers. That folded
into our voice models will help. We usually recognize UK English fairly well.
I'll have to send a note to the IBMers in Australia using the product and ask
for their results.

Best Regards,
Ken Solkowitz
IBM Continuous Speech Series
Bethesda, MD 20817
301-803-3897



#: 904 S10/IBM Speech Products ∙5 ∙
    20-Feb-94  07:20:35
Sb: #874-Other Language Lexicons
Fm: Michael Burgun 100020,1604
To: Ken Solkowitz 72103,2111

Ken,

I doubt you will have any trouble with Australians - I was just "stirring"
(Australian for teasing).

Michael



#: 882 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:22:51
Sb: #748-Other Language Lexicons
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Hutchison 72254,1315

Hi Bob

I don't find American accents change that much (I've lived in New York State,
Maryland, and Oregon, and traveled to much of the rest of the country).  I
don't know of any American accent that ICSS can't handle.  At trade shows we
have seen ICSS work with various foreign accents.

Brandon

ICSS Development - IBM

#: 1 S10/Thread Header
    16-Feb-94  16:27:26
Sb: #378-An Odd Application
Fm:
To:

TITLE: AN ODD APPLICATION
KEYWORDS: TIMING TIMESTAMP START

#: 819 S10/IBM Speech Products ∙5 ∙
    16-Feb-94  16:27:26
Sb: #378-An Odd Application
Fm: Rufus Laggren 76346,1441
To: Brandon Booth [IBM] 71042,3144

 >> The way ICSS works your program tells it when to start gathering
 >> speech data and when ICSS returns with the reco'd text it also returns
 >> timing data (utterance length, reco time).  With this your program should
 >> be able to coordinate it's input sources.  What you will not get from ICSS
 >> is timing information on a word by word basis.

Can ICSS emit msg's to the opsys at pre-specified times (such as "start of
sound" or "end of sound")?  Or can it "timestamp" its sound *and* text stream
at predefined intervals?

I'm assuming (from complete ignorance) that it emits at least 2 outputs, the
text and sampled sound stream (?).

Rufus

#: 888 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:23:20
Sb: #819-An Odd Application
Fm: Brandon Booth [IBM] 71042,3144
To: Rufus Laggren 76346,1441

Rufus

ICSS does not currently post messages for Start of Talking, etc.  We have had
this requirement from other customers and it is under consideration.  And yes,
we do return the reco'd text, and you can get the sound stream back if you
want it.

Brandon

ICSS Development - IBM

#: 921 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  01:55:03
Sb: #888-An Odd Application
Fm: Rufus Laggren 76346,1441
To: Brandon Booth [IBM] 71042,3144

 >> ICSS does not currently post messages for Start of Talking, etc.  We
 >> have had this requirement from other customers

Yes, it seems useful to capture a few "milestones" to maintain a timeline, if
it's not too expensive.  Thnks for you quick note.

#: 1 S10/Thread Header
    16-Feb-94  19:05:50
Sb: M-ACPA vs SoundBlaster
Fm:
To:

TITLE: M-ACPA VS SOUNDBLASTER
KEYWORDS: M-ACPA MWAVE M-WAVE SOUNDBLASTER SB SOUNDBOARDS

#: 826 S10/IBM Speech Products ∙5 ∙
    16-Feb-94  19:05:50
Sb: M-ACPA vs SoundBlaster
Fm: Vernon Harvey 72120,2767
To:  Brandon Booth (IBM) 71042,3144

Brandon:

I am currently using the ICSS software with the MMPM/2 and a SoundBlaster 16
ASP card.  I have just discovered that the developer's toolkit only allows me
to write PM applications when using the "icssmmpm" device, rather than being
able to write simple text mode C programs.  This is not a big deal, but slows
down prototyping a bit.

As a result, I am curious.  What are the pros and cons of the SoundBlaster
versus the M-ACPA card.  Are there any features on the SB card that I cannot
get on the M-ACPA card or vice versa.  And most importantly, will there be any
difference in voice recognition by using one card or the other.

I have never seen the M-ACPA card advertised anywhere, so any pricing info and
features lists would be more than welcome.

Thanks again.
Vernon Harvey.

#: 890 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:23:28
Sb: #826-M-ACPA vs SoundBlaster
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Hi Vernon

Well the basic difference between SB and ACPA is that SB's are very popular
and ACPA's nearly unknown.  In fact IBM is converting over to another series
of cards based on MWAVE technology.  We have not noticed any difference in
reco accuracy between the two cards (16 bit in the case of SB, the 8 bit cards
do reco worse).

Brandon

ICSS Development

#: 939 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  14:51:56
Sb: #890-M-ACPA vs SoundBlaster
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

Thanks for the info on the differences.  I guess I will stick with the SB card
for now, and maybe get the M-Wave card when it is released for OS/2.  Is the
M-Wave card the one that incorporates voice, data, fax and sound card in a
single board.  I remember seeing something at the PACRIM show here, and it
sounded pretty impressive, except for the 9600 baud modem (Personally, I want
V.Fast, but I will settle for 14.4).  The card I saw however, only had Windows
 software, and they said that an OS/2 version was in the works.  Any idea when
its due out?

Regards,
Vernon.

#: 953 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  20:24:41
Sb: #939-M-ACPA vs SoundBlaster
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Vernon

I'm sorry, I don't know much about the M-Wave based cards.  You might ask over
on the MMPM section of OS2DF1.

Brandon

ICSS Development

#: 1 S10/Thread Header
    17-Feb-94  17:55:09
Sb: ICSS - H/W & S/W Quests
Fm:
To:

TITLE: ICSS - H/W & S/W QUESTS
KEYWORDS: TIMING TIMESTAMP START STORAGE DIGITIZE

#: 843 S10/IBM Speech Products ∙5 ∙
    17-Feb-94  17:55:09
Sb: ICSS - H/W & S/W Quests
Fm: Marc S Grushcow 70751,1501
To: Brandon Booth 71042,3144

Brandon,

It looks like we might have a paying project that will let us use both OS/2
and ICSS.  It will be an interesting way to get our feet wet.  I have some
questions about the ICSS API and hardware...

1) Timing
I want to know when the speaker started an utterance.    ICSSGetSpokenWords
returns AcquisitionTime and RecognitionTime.   Assume that the function just
completed and I immediately get the system time.  How good an estimate of
utterance start time is system time minus the sum of the acquisition and
recognition time?

2) Storage
Is it possible to store the digitized data as well as having it go to ICSS for
processing.  Ideally, I'd like ICSS to give me the digitized data along with
the text, but I can't find any support for this in the manual.  If it could do
this, I wouldn't have to save "silence",  but I would happily settle for a
complete record of the digitized session (yes I know the sample rate is
22KHz).

3)Processor, memory and disk
We are looking at a 48xDX2 at 66MHz with 16 or 24MB RAM which meets the h/w
requirement.  At 16MB, would we expect to see swapping?  I don't expect tht
our app would be really big. (I know that that isn't a lot of help).  In any
case, we will probably go SCSI for the disk (.5 or 1GB)  and are thinking
about the Adaptec local bus controller (2742 I think).  Any comments?  Again,
we would like to be able to save the digitized data in parallel with speech
analysis.

4)Sound Cards
I gather that ICSS still doesn't support any DSP units.  We will probably get
the Pro Audio Spectrum 16 unless there is a technical reason to go with the
M-ACPA card.

5)A Grammar Question
We will probably go with bigrams based on sample sentences and sentence
fragments.  We have quite a lot of these.  I was wondering about the BNF form
though.  If I write a BNF that can handle...
        Buy grapes, oranges, cheese, and bread and beer.
will it also handle...
  Buy grapes, oranges ...<long pause>... cheesse, and bread and beer.
where <long pause> is longer than Minimum Silence Time?

Thanks in advance

Marc

#: 848 S10/IBM Speech Products ∙5 ∙
    17-Feb-94  22:11:37
Sb: #843-ICSS - H/W & S/W Quests
Fm: Bob Cave 73232,2474
To: Marc S Grushcow 70751,1501

Marc,

Hi.  I am sitting in for Brandon for a few days, and I think that I can answer
your questions:

1) Timing - the system time minus sum of acquisition time and reco time should
produce a pretty good estimate.  How close does the application need to be?

2) Storing digitized data - No problem here.  You can tell ICSS to record each
utterance while it performs recognition.  Check the system parameter called
Record Session Name.  You give it a path and file name, and ICSS records each
utterance in a separate file with .001, .002, etc. appended to the file name.
I have used this several times, it works great.

3) Hardware reqts - your configuration is quite similar to mine back at the
office.  I do get swapping with 16 MB RAM, but I also tend to have a bunch of
other things running at the same time.  SCSI disk with the Adaptec controller
sounds great...I wish I had one!  If you want to save 22khz samples, go with
the 1GB drive.

4) Sound Cards - PAS 16 is fine, and you are correct--no DSP support.

5) Grammar - when the user is silent for a time greater than the current value
for end talking silence, ICSS will stop listening.  You can increase this
value, but that will make all recognition response times longer.

I hope this helps.

Bob Cave, ICSS Development Team

#: 891 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:23:37
Sb: #843-ICSS - H/W & S/W Quests
Fm: Brandon Booth [IBM] 71042,3144
To: Marc S Grushcow 70751,1501

Hi Marc

On your first question, to be honest I'm not sure how good a number you will
get.  Our acquistion time starts when we start acquiring data (hence the
name).  The start of talking may occur well after that.  We've had customers
ask for messages to be posted back on these kinds of events, and we are
evaluating this enhancement.

On saving the speech signal, look at the SetValue function and the
RecordSpeech and RecordSession settings.  That will do what you want.  And you
can use either the 22 kHz or 11 kHz sample rates.

16 meg main ram should not show any swapping, so the disk (which sounds quite
nice) shouldn't factor in (except of course if you are saving the sampled
data).

ICSS does not off load DSP processing, so the PAS card is a fine choice.

In any case if a user pauses longer than the end silence time, ICSS will think
she has ended the utterance and will begin to process the speech data.  You
can play with this value to get a balance that makes sense to you.

Glad I can help.

Brandon

ICSS Development - IBM

#: 929 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  09:12:53
Sb: #891-ICSS - H/W & S/W Quests
Fm: Marc S Grushcow 70751,1501
To: Brandon Booth [IBM] 71042,3144

Brandon,

I think I liked Bob's answer about identifying the start of speech time better
than yours.  I hate to start an argument, but who is right?  This could be a
sticky problem for us.

Our project is dealing with multimodal dialogues - we are getting input from
both voice and gestures (pen, touch screen or mouse).  We can save the "ink"
from the gestures and timestamp them for later analysis.  It looks like speech
is going to be a problem.

Two things... First we want to know when the utterance occured (start time and
duration).  Second, I think that we want to have the whole voice sessions
recorded, not just the slices where ICSS thinks that it found something.  The
scientists involved will want to be able to playback the digitized voice and
see where ICSS found things at least until they are confident that it is
working.

??
Marc



#: 940 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  15:04:17
Sb: #929-ICSS - H/W & S/W Quests
Fm: Vernon Harvey 72120,2767
To: Marc S Grushcow 70751,1501

Hi Mark:

Your system sounds a bit similar to my home system on the H/W level.  You
might want to check the availability of OS/2 2.1 drivers for the Adaptec VLB
controller.  When I bought my system in December, Adaptec had the VLB
controller, but no driver support for OS/2.  As a result, I went with the
BusLogic BT-445S (I think that's the right model), and it is working without
any problems.  Adaptec was supposed to have drivers
available sometime in January.

Regards,
Vernon Harvey.

#: 951 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  20:24:33
Sb: #929-ICSS - H/W & S/W Quests
Fm: Brandon Booth [IBM] 71042,3144
To: Marc S Grushcow 70751,1501

Hi Marc

I'm sorry, but I believe I am correct about the timings returned.  Let me make
a destinction between the entire audio data and the utterance data, which is
from the start of talking (when the speech signal rises above the threshold)
to the end of talking (when the speech signal falls below the threshold).  In
the shipping products we return the acquistion time, which is the first of
these.  Customers have requested being notified of the events in the second of
these.  We are evaluating this request.  Also,  you can record all of the
audio data from the time the card is opened until the time it is closed, and
analyse this if you wish.

Brandon

ICSS Development

#: 983 S10/IBM Speech Products ∙5 ∙
    22-Feb-94  21:13:53
Sb: #929-ICSS - H/W & S/W Quests
Fm: Bob Cave 73232,2474
To: Marc S Grushcow 70751,1501 (X)

Marc,

Sorry, but I must yield to Brandon on this one.  He is much closer to that
part of the code than I am.

Bob

#: 1 S10/Thread Header
    03-Feb-94  00:57:24
Sb: ICSS problem
Fm:
To:

TITLE: ICSS PROBLEM
KEYWORDS: WINDOWS INSTALLVIRTUAL MEMORY SWAPFILE BANDWIDTH MICROPHONE WIN32

#: 634 S10/IBM Speech Products ∙5 ∙
    03-Feb-94  00:57:24
Sb: ICSS problem
Fm: Harry R. Crowe 72677,2714
To: sysop

I have just received the ICSS package for Windows v3.1 My install failed.  The
process accepted all 9 disks, and then reported that I did not have Virtual
Memory activated. My environment is
  MSDOS v6.2
  PC = Clone 486DX2/66 with 20M installed RAM
  Disks are C=170M and D=130M
  Memory management is via native mode MS drivers
    (himem.sys and emm386.exe)
  Chosen installation area is D-drive with 40M free When I tried to access the
test program (Developer's Toolkit) the system responded by
  NO load
  NO action
  RETURN to Program Manager Nothing else in the ICSS window worked properly.
The DOS path had been ammended and re-boot was performed. I then blew away the
installed programs on D: and attempted to re-install the system.  This was not
smoothe.  The same result occurred, and I have now given up (in disgust).....
Can you help me???????

#: 642 S10/IBM Speech Products ∙5 ∙
    03-Feb-94  15:46:43
Sb: #634-ICSS problem
Fm: Ora Williamson 71742,2161
To: Harry R. Crowe 72677,2714

>>I have just received the ICSS package for Windows v3.1 My install failed.

Harry:

Did you restart Windows before trying anything?

In the Control Panel under 386 Enhanced, check your Virtual Memory size.  Can
your swap file grow to 12 Meg or larger?  If not, set it up as thus, and try,
try again.

Don't get upset!  Once you are correctly configured, you'll be very happy!

Regards, Ora Williamson IBM Spoken Language Systems Lab

#: 671 S10/IBM Speech Products ∙5 ∙
    05-Feb-94  22:15:02
Sb: #642-ICSS problem
Fm: Harry R. Crowe 72677,2714
To: Ora Williamson 71742,2161

Thank you for your response.  You proposed two steps to start my path to
success
with ICSS:
 1.  Restart Windows
 2.  Generate a 12M swapfile
I honestly do not remember if I first restarted windows, but now any time I go
into the Developers Toolkit the screen blackens momentarily and then returns
to the Program Manager.  I have not yet implemented the 12M Swap File.  I will
do so now with my new 345M hard disk and see what happens - remember from my
first message:  I have 20M (!) of installed RAM.  Do I still need such a swap
file???
I may be checking back!
thanks again...
HC

#: 673 S10/IBM Speech Products ∙5 ∙
    06-Feb-94  01:07:36
Sb: #642-ICSS problem
Fm: Harry R. Crowe 72677,2714
To: Ora Williamson 71742,2161

Ora:
 This is one of those times when my extensive computer background is feeling
like a light shirt on a cold night...  I have messaged you once since my first
call for help regarding ICSS for Windows.  This email is before any reply to
my second message would be possible for you (I think).  At your suggestion, I
created a >12M swap file.  But I also re-installed the system to my C-drive
(instead of D).  The install went more smoothely, and I can now bring up the
Developer's Toolkit.  But I cannot DO anything.  If I attempt to TEST the
system, I get various non-useful responses of the type that make one crazy.
The TechRef manual says to go to the DevToolKit and select Context from the
Actions pull-down menus....  There IS no Context Menu Item!!  As a matter of
fact, the manual does not seem to correspond to any screens that I have
intially found!  This is most frustrating.  By sheer stubborn menu-hunting I
have found a Grammars/BNF/File/Context  item that seems to locate the Testing
context.  But then God help me if I try to do anything with it (such as
ACTIVATE it, as stated in the manual).  HELP!  I only have one life and do not
wish to spend it on ICSS-in-the-dark!  Do I have the wrong manual for the
software I was sent???...
For reference, my environment is as follows.  I have a 486dx2/66 with MSDOS
v6.2, which performs other multi-media operations fine.  The sound card is a
Sound Blaster 16/ASP that DOES record in SB software - so I know the hardware
works...  Any assistance would be greatly appreciated.  I can call on a voice
line during working hours, but the machine is at home and you know how hard it
is to trouble shoot something you are not in front of!
Thanks,
HC



#: 682 S10/IBM Speech Products ∙5 ∙
    06-Feb-94  22:38:14
Sb: #673-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Harry R. Crowe 72677,2714

Harry

When Ora was speaking about testing ICSS, she was refering to the Test Context
program (wtstctx.exe).  This can be invoked from the Development Environment
panel from the Speech pulldown (I think this is correct, I don't quite
remember).  You need to load a compiled context that is shipped with ICSS.
These have extension .ctx.  Please make sure you have the ICSSDIR environment
variable set to the install directory of ICSS (if its installed on the C:
drive than ICSSDIR should be set to C:\ICSSWIN).  The Grammar pulldown is for
compiling contexts.  Your machine config sound fine.  Please make sure that
you can record and play back audio with your setup.

Brandon

ICSS Development

#: 688 S10/IBM Speech Products ∙5 ∙
    07-Feb-94  16:35:46
Sb: #673-ICSS problem
Fm: Dan Devine 76040,215
To: Harry R. Crowe 72677,2714

Harry,

I was getting a similar problem when I tried to activate the testing context
(see my
other post).  I spoke to tech support today and they said I was getting the
error
because I only loaded the High not the Medium.  The included contexts were
created with the Medium so I need to install that as well.  This was not
stated in
the manual either.  Hope this helps.

Dan



#: 694 S10/IBM Speech Products ∙5 ∙
    07-Feb-94  22:59:00
Sb: #688-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Dan Devine 76040,215

Dan is quite correct.  The ICSS contexts are input sample rate specific.  We
have two input rates: High(22050 kHz) and Medium (11025 kHz).  The contexts
that we ship are compiled in the Medium rate (bandwidth).  You need to have
the Medium models installed in order to do reco with a Medium context.

Brandon

ICSS Development



#: 710 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  00:37:45
Sb: #694-ICSS problem
Fm: Harry R. Crowe 72677,2714
To: Brandon Booth [IBM] 71042,3144

BB:
Thank you for your response.  With Ora's and your help, the system now seems
to be up and running!  I re-installed with the Medium format and the
Test/Sample operations worked fine....  By the way, just what does the WIN32S
component provide?  I took a removable (ICSS installed) disk from home to
work, and the DevToolKit would not come up claiming (in one of those fast,
dark screen blips before going back to the program manager that "This program
does not work in DOS mode").  Is this because I had not done an INSTALL on the
work computer?  Should I do an incremental install of the WIN32S component to
funtion on both?  This is a matter of curiosity, but I WOULD like to know!
Also, while perusing the HELP documents, I noticed a query regarding other
processors.  Is the MathCo absolutely necessary (I have a 486dx2/66 at home,
so I am OK there)??  Will a 386/40 work by itself?  And if not would it with a
MathCo?
Thanks for your input...
HC

#: 734 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  21:16:18
Sb: #710-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Harry R. Crowe 72677,2714

Hi Harry

Glad you got running.  Ora is an amazing help to a lot of people (she's a lot
more people oriented than I; I'm more of a like to program in a cave sort). We
use the Microsoft Win32s support to provide linear memory (as opposed to
segmented memory normal to Windows 3.1), and this is mostly because we ported
from OS/2.  ICSS will not work unless Win32s support is installed (although
some components will come up, as they don't use Win32s).  Part of ICSS heavily
uses floating point operations and the program is computationally intensive
enough that a 486 DX 33 mHz machine is the smallest machine I'd recommend.  A
486 DX2 66 mHz is a very nice machine for ICSS (I develop on one at work - a
clone actually).  Did I answer all of your questions?

Brandon

ICSS Development


#: 747 S10/IBM Speech Products ∙5 ∙
    10-Feb-94  12:27:00
Sb: #734-ICSS problem
Fm: Bob Hutchison 72254,1315
To: Brandon Booth [IBM] 71042,3144

Brandon,

I purchased ICSS for windows unaware that it required win32s.  Now, this
wouldn't normally be a problem, but for some reason win32s will not run with
my windows drivers (S3) -- really annoying, but not your problem.

Why do you use win32s?  If it is for 32bit support you can get that with
Watcom's compiler and you would be able to run on just about any windows
machine.

Bob

#: 881 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:22:47
Sb: #747-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Hutchison 72254,1315

Bob

We use Win32s for the flat memory addressing as we ported from our OS/2 2.1
version.  We felt it was the best solution.

Brandon

ICSS Development - IBM

#: 932 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  10:49:58
Sb: #881-ICSS problem
Fm: Bob Hutchison 72254,1315
To: Brandon Booth [IBM] 71042,3144

Brandon,

I guess the cost to me is that I cannot develop for OS/2 and Windows at the
same time.  Unless you know something I don't about Win32s under OS/2.

Bob

#: 952 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  20:24:38
Sb: #932-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Hutchison 72254,1315

Bob

If it helps, developing ICSS for all of the operating systems that we do is as
much (or more) of a headache.  I'm sorry we can help.

Brandon

ICSS Development

#: 961 S10/IBM Speech Products ∙5 ∙
    22-Feb-94  01:03:03
Sb: #952-ICSS problem
Fm: Richard Frank 73511,3406
To: Brandon Booth [IBM] 71042,3144

Brandon,

  Is a Sony F-VX500 mic good enough to use?  I have no frame of reference yet.

  RIchard

#: 989 S10/IBM Speech Products ∙5 ∙
    22-Feb-94  23:34:20
Sb: #961-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Richard Frank 73511,3406 (X)

Richard

I don't know.  We've used the mics listed in the tech ref (Sennheiser 410,
518, 431; Conneaut Audio Devices 89; Electro-Voice 857; Radio Shack 33-1037,
33-1067).  If you have good results, let us know.

Brandon

ICSS Development

#: 1008 S10/IBM Speech Products ∙5 ∙
    23-Feb-94  15:50:25
Sb: #952-ICSS problem
Fm: Bob Hutchison 72254,1315
To: Brandon Booth [IBM] 71042,3144

Brandon,

I'm trying not to criticise too much :-)  I appreciate the work you are faced
with.

Bob



#: 711 S10/IBM Speech Products ∙5 ∙
    09-Feb-94  00:37:47
Sb: #688-ICSS problem
Fm: Harry R. Crowe 72677,2714
To: Dan Devine 76040,215

Dan:
Thanks for your input.  I was in the same boat:  "High Bandwidth only"...  I
then  installed the Medium and all was fine!  This area is exploration for me,
but what are people (e.g. you) doing with this??  It is obviously something
that would be served by a better version of OLE!?  Like Interactive Object
Embedding?  The program development route is pretty long, with many
corners!...
Anyway thanks for the response...
HC



#: 669 S10/IBM Speech Products ∙5 ∙
    05-Feb-94  21:58:46
Sb: ICSS problem
Fm: Terry Jones 73131,3137
To: sysop

I appear to have a couple of problems with the Windows development kit I
bought and hope you can help.

When I was installing the software everything went fine up to disk 9, the
WIN32S diskette. When it started up it looked at the A: drive despite all the
others were loaded on the B: drive, and it wasn't possible to load it. How do
I load this?

The second problem is that when I followed instructions on testing the ICSS
installation I got the error

"StartConversation failed rc:15, Invalid ADCIF Number has been
        specified"

after selecting TEST CONTEXT in SPEECH. I tried the stock exchange demo and I
got the following error:

"Invalid ADCIF Number has been specified"

Any idea what is causing this, and how to solve it?

My system is a 486 clone, I am using a SoundBlaster board and am using QEMM.

Hope you can help,


Terry Jones

[phone]



#: 681 S10/IBM Speech Products ∙5 ∙
    06-Feb-94  22:38:08
Sb: #669-ICSS problem
Fm: Brandon Booth [IBM] 71042,3144
To: Terry Jones 73131,3137

Terry

You have to have the Win32s support installed for ICSS to work.  I don't know
why the Win32s disk was looking for the A drive.  What you should do is call
our bug report number (800 - 627 - 8822) and describe your problem to Dan (the
guy who answers that line - he's good at it).  The ADCIF message means that it
(ICSS) thinks you have the wrong card number (this is set in one of the
profiles - see the info in the tech ref on profiles).  If you have a single
audio card installed, I believe 0 is the card number to use.    However, I
would distrust any error messages until you have Win32s correctly installed.

Brandon

ICSS Development

#: 1 S10/Thread Header
    28-Feb-94  09:34:24
Sb: ICSS questions
Fm:
To:

TITLE: ICSS QUESTIONS
KEYWORDS: VISUAL BASIC VBX CONFIDENCE CERTAINTY SPEED BANDWIDTH

#: 1094 S10/IBM Speech Products ∙5!∙
    28-Feb-94  09:34:24
Sb: ICSS questions
Fm: Stephen Ng 70302,1510
To: all

I've been playing with ICSS/Windows and have been quite impressed with it.  I
have a couple of comments/questions/suggestions:

1. I got ICSS working with Visual Basic; it wasn't too bad except I had to
write 2 small C routines to do type conversions, and I also had to use the
MessageBlaster VBX to handle the ICSS reply messages.  Of course what would
really be nice here would be a VBX--would anyone else be interested in such a
creature?

2. I hooked up ICSS to my telephone, and have been getting decent results.  I
have a little restaurant locator app running on a Dialogic board, and you can
select from among 20 different cuisines by speaking the name of the cuisine
(French, Italian, etc.)  (617-643-EATS in case you want to try it.)  It seems
to have the most trouble recognizing single syllable words (it always wants to
interpret "French" as "Greek").  The radio shack gadget for getting a
mic-level output from the phone seems pretty cheap and may be adding some
distortion.  I also need to fool with the sound card mixer settings to make
sure volume is optimal.  Nevertheless, what about adding some kind of
certainty rating to the API so I can see how good the match is?  And also a
way to get the next-closest-guess so I can ask the user about the choices?

3. I know I'm really pushing ICSS, since it's designed to work with
high-quality microphones directly into a sound card, and not the telephone
handsets over phone lines.  But I'm trying to make things easier by keeping
the grammars trivial, and the vocabularies small.  Is it possible to adjust
sound model for typical telephone handset acoustics by using a graphic
equalizer or doing some signal processing on the sound?  Can I adjust the data
tables ICSS uses?  Or, since it seems to have a propensity to think that most
single-syllable cuisines are "greek" I can bias the bigram table so that
"greek" is less likely?  (I can upload the .bnf/.ctx/.wav files if it would
help.)

4. I read that ICSS may use the first utterance to adjust for the particular
mike.  Does that mean the first utterance since the ICSS was started?  Or per
conversation?  My application uses a single conversation for multiple phone
calls--should I use different conversations for different calls so it can
readjust for different phones?

5. Perhaps ICSS has a hard time distinguishing the vowels in "French/Greek".
So would it help if I changed my grammar such that the user had to say
"please" after the cuisine name?  I.e. "French, please?"  ICSS would know that
the second word was "please", and perhaps that could help it figure out that
the vowel in "french" was not the same sound???

6. The 1000-word "limit"--does the performance degrade linearly or
exponentially as we approach this limit?  Another way to ask the question is,
what if I got a machine twice as fast as my 486/66--could I recognize 2000
words as quickly as 1000 words now?

Thanks for your help,

--Steve



 #: 1106 S10/IBM Speech Products ∙5!∙
    28-Feb-94  21:05:44
Sb: #1094-ICSS questions
Fm: Ken Solkowitz 72103,2111
To: Stephen Ng 70302,1510

Stephen -
I am going to bring your post into work and have Bob take a look at it.
Knowing Bob, he'll take care of this tonight making tomorrow's discussion a
mute point.

Regards,
Ken Solkowitz
IBM Continuous Speech Series
Bethesda, MD 20817
301-803-3897

#: 1111 S10/IBM Speech Products ∙5!∙
    28-Feb-94  23:41:42
Sb: #1094-ICSS questions
Fm: Bob Cave 73232,2474
To: Stephen Ng 70302,1510

Stephen,

That is quite a post!  I can answer some of your questions now, others will
require some consultation "back at the ranch".

>1)...the ICSS reply messages.  Of course what would really be nice here would
>be a VBX--would anyone else be interested in such a creature?

Given some of the discussion in this topic over the past couple of days, I
think the answer is definitely "yes".

2) We are investigating the notion of adding some type of "confidence"
indicator to the response for a recognition.  I am not sure about returning
the next best response...I will look into it.

>3. I know I'm really pushing ICSS, since it's designed to work with
high-quality >microphones directly into a sound card, and not the telephone
handsets over >phone lines.

This is true: you are pushing ICSS quite hard, possibly beyond its
capabilities.  What is your hardware configuration?  How are you getting voice
data into ICSS?  Are you splitting the phone line, with one end going into the
Dialogic card and the other end into a sound card?  The difficulties with
using ICSS over the phone come from the reduced bandwidth present in the sound
samples.  Our speech models are tuned for microphone input bandwidth.  I don't
think an equalizer will help much.  The models ("data tables") cannot be
modified by an application.  In general, the longer and more distinct your
words are, the better recognition you will get.

>4. I read that ICSS may use the first utterance to adjust for the particular
mike.

This is true, and the adjustment is per StartConversation call, I believe.
Using a new conversation for each call would add overhead to your calls.  You
can minimize this by making sure that there is always at least one context
loaded.  The initial LoadContext call causes a large file called a Common
Probability Distribution file to be loaded, and this can take several seconds.
  Once it is loaded, all contexts use it and subsequent LoadContext calls are
quite fast.

>5. Perhaps ICSS has a hard time distinguishing the vowels in "French/Greek".
>So would it help if I changed my grammar such that the user had to say
"please" >after the cuisine name?

I do not think this will buy you anything, but I will check it out.

>6. The 1000-word "limit"--does the performance degrade linearly or
exponentially >as we approach this limit?

Performance for ICSS recognition is affected more by the branching factor
(number of choices for the next word at any given point within a grammar)
within a context than the number of different words.  In any case, a faster
machine will definitely produce faster recognition, but not necessarily more
words.

I hope this helps.  Please post back if you have any more questions.

Bob Cave, ICSS Development

#: 1155 S10/IBM Speech Products ∙5!∙
    02-Mar-94  10:58:44
Sb: #1111-ICSS questions
Fm: Stephen Ng 70302,1510
To: Bob Cave 73232,2474

Bob,

Thanks for your very helpful response!  I look forward to any more info you
may have about my questions; meanwhile here are some responses to some of your
questions:

You write:
"Given some of the discussion in this topic over the past couple of days, I
think the answer is definitely "yes"."

I will look into developing a VBX interface to ICSS and possibly distributing
it as shareware.  (Would anyone care to volunteer how much this would be worth
to them?  "IBM should provide it for free" is a valid, though not encouraging
to me, response.)

You write:
"What is your hardware configuration?  How are you getting voice data into
ICSS?  Are you splitting the phone line, with one end going into the Dialogic
card and the other end into a sound card?  The difficulties with using ICSS
over the phone come from the reduced bandwidth present in the sound samples.
"

I'm running a 486/66DX2 with 16 meg under Windows 3.1, using a Pro Audio
Studio 16, but I'm certainly open to using upgrading my hardware if it will
help.  Yes, I'm splitting the phone line just as you say.  I realize this is a
difficult problem, but I am hopeful that ICSS will provide at least as good a
solution as some of the dedicated hardware voice recognition solutions
designed for just telephones (speaker-independent, discrete speech, small
vocabulary).

You write:
"In general, the longer and more distinct your words are, the better
recognition you will get."

That has definitely been my experience; my biggest problem is with recognizing
a single syllable word.

You write:
">4. I read that ICSS may use the first utterance to adjust for the particular
mike.
This is true, and the adjustment is per StartConversation call, I believe."

Interesting, so perhaps at the start of each phone call, the I should do a
StartConversation, and ask the user to say "hello"?

You write:
"Performance for ICSS recognition is affected more by the branching factor
(number of choices for the next word at any given point within a grammar)
within a context than the number of different words.  In any case, a faster
machine will definitely produce faster recognition, but not necessarily more
words."

So I should keep the grammar perplexity down if ICSS gets too slow.  But what
if I have a trivial grammar?  I don't understand your comment that a faster
machine will not buy me more words....the manual says that 1000 words is not a
hard limit, but that I simply won't like the performance as I approach 1000
words.  Doesn't that mean that I can go over 1000 with a really really fast
machine?

Thanks,


Steve

#: 1163 S10/IBM Speech Products ∙5!∙
    02-Mar-94  17:18:52
Sb: #1155-ICSS questions
Fm: Vernon Harvey 72120,2767
To: Stephen Ng 70302,1510

Stephen:

I seem to recall when I was playing with the Dialogic voice card, that there
was an interface bus which allows you to take the audio signal directly off
the card and feed it to another card.  If your audio card has an input
connector, maybe you could patch them together to provide a cleaner audio
connection, possibly with less signal loss.

Regards,
Vernon.

#: 1172 S10/IBM Speech Products ∙5!∙
    02-Mar-94  21:48:04
Sb: #1163-ICSS questions
Fm: Bob Cave 73232,2474
To: Vernon Harvey 72120,2767

Vernon,

The interface bus used by the Dialogic cards is proprietary, although that is
opening up with the new Signal Computing Systems Archtecture (SCSA) standards.
One can, however, use a Dialogic matrix switch card to provide input to the
sound card through its audio input jack.  (I wish I had though to mention this
in my previous reply to Steve <g>).

Thanks,

Bob Cave, ICSS Developoment

#: 1182 S10/IBM Speech Products ∙5!∙
    03-Mar-94  13:44:11
Sb: #1155-ICSS questions
Fm: Vernon Harvey 72120,2767
To: Stephen Ng 70302,1510 (X)

Stephen:

I did a quick look through my technical info on the Dialogic card last night.
If you have a D/4X series card, there is a 20 pin bus connector which allows
you to get audio from the card for any of the 4 channels.  I found the info in
the appendix, and it may be of some use to connect to card to your sound card
for speech recognition.  The connector is referred to as a P3 audio connection
bus in my Dialogic manual.  I don't think that you need the Dialogic Matrix
switch in order to connect the two cards, however, you may want to check with
Dialogic tech support to be sure.

Regards,
Vernon.

#: 1333 S10/IBM Speech Products ∙5!∙
    11-Mar-94  11:03:39
Sb: #1182-ICSS questions
Fm: Stephen Ng 70302,1510
To: Vernon Harvey 72120,2767

Vernon,

Thanks for the tip!  What publication are you looking at?  I've looked my D/4x
Hardware reference book, but it seems to just refer me to other docs.  I'll
try to get the info from Dialogic directly....  I would like to try hooking up
to the bus directly to see if I can get a cleaner signal.  (The main issue
appears to be making sure my sound card doesn't get fried when the phone
rings.)

--Steve

#: 1344 S10/IBM Speech Products ∙5!∙
    11-Mar-94  18:06:11
Sb: #1333-ICSS questions
Fm: Vernon Harvey 72120,2767
To: Stephen Ng 70302,1510

Stephen:

Sorry for the delay in replying, I have not been online lately (Busy cruising
around in my new car!).   The manual which I found this information in was the
Dialog/4X User's Guide, version 2.30, which appears to date back to about
1988.  It was in one of the appendices.  I have come to the point where I
never throw out the old manuals, as it seems that there are always ommissions
in the future releases.  Particularily when the company decides that it can
make more money selling the information.

Glad to have been able to help.

Regards,
Vernon Harvey
Rydex Industries Corporation.

#: 1242 S10/IBM Speech Products ∙5!∙
    06-Mar-94  18:20:14
Sb: #1155-ICSS questions
Fm: Rufus Laggren 76346,1441
To: Stephen Ng 70302,1510

>... Shareware VBX for ICSS...  value ??

IMHO:

Workable -> $30+

Very good -> $50+

Just a guess, I'm afraid, and biased to the low-medium end.  I.e., if the base
package becomes available for about $200 or less (hopefully), it allows
"impulse purchase" experimentation by many developers and a VBX priced as
above would fit that market.  However, I confess to usually being low in my $$
estimates.

Would suggest that you include a nag screen (not too gross), and configure
your release to allow some material enhancement to be provide through
registration.  E.g., distribute a fully usable package, but provide optimized
dll's or some small useful enhancements with registration...  I judge others
by myself, and I think that such a "reminder" is needed in shareware, and that
although it's a departure from the pure free test traditionally provided, it's
justified and fair and ultimately acceptable.

I'd like to shareware continue to produce fine products, and I think that many
people which otherwise would just use the product forever free would consider
registering  to gain a some small-but-not-purely-symbolic enhancement if they
indeed find the base product fairly useful.

Hope you find it do-able.  Good luck.

#: 1243 S10/IBM Speech Products ∙5!∙
    06-Mar-94  18:20:21
Sb: #1111-ICSS questions
Fm: Rufus Laggren 76346,1441
To: Bob Cave 73232,2474

 >> The difficulties with using ICSS over the phone come from the reduced
 >> bandwidth present in the sound samples.  Our speech models are tuned for
 >> microphone input bandwidth.

Is it possible to utilize different "speech models"?  Perhaps one optimized
for telephony?   Is it feasible to create a speech model through a training
session using the actual equipment with a specific limited speech/text stream
which gets recorded and then corrected or "set" into the speech model?  Note
opportunity for additonal products here <g> (various speeck models,
"mastering" software to create speech models...).

 >> Performance for ICSS recognition is affected more by the branching
 >> factor (number of choices for the next word at any given point within a
 >> grammar) within a context than the number of different words.

Hmm.  Accruacy depends on the "number of choices for the next word..", but not
the "..number of different words"? .....  ??????

Can I *preset* the branching profile for the next incoming speech stream *and*
preset the available vocalbulary selection such that amoung very limited
choices the hit-rate or confidence level rises substantially?  Or is this
already an automatic part of the ICSS functionality?

No rush.  When you get a chance.  Thanks.

#: 1245 S10/IBM Speech Products ∙5!∙
    06-Mar-94  21:07:13
Sb: #1243-ICSS questions
Fm: Bob Cave 73232,2474
To: Rufus Laggren 76346,1441

Rufus,

 >>Is it possible to utilize different "speech models"?
 >>Perhaps one optimized for telephony?

Yes, this is possible, and I agree with regard to the additional product
opportunities.  Many of our customers (and potential customers) have expressed
a requirement for a version of ICSS which works in a telephony environment.

 >>Hmm.  Accruacy depends on the "number of choices for the next word..",
 >>but not the "..number of different words"? .....  ??????

When I said "performance" in the quote you included, I was referring to
recognition speed, not recognition accuracy.  I should have been more
explicit.

>> Can I *preset* the branching profile for the next incoming speech stream
>>*and* preset the available vocalbulary selection such that amoung very
>> limited choices the hit-rate or confidence level rises substantially?  Or
is
>>this already an automatic part of the ICSS functionality?

You can have multiple vocabularies (or "contexts" in ICSS terminology) loaded
into memory at any given time.  Each time you perform recognition, you can
select any one of the loaded contexts.  You will get the best accuracy by
defining small contexts with words which are distinct from one another rather
than putting all possible responses into a single context.  Does this answer
the question?

Regards,

Bob Cave, ICSS Development


#: 1259 S10/IBM Speech Products ∙5!∙
    07-Mar-94  23:39:08
Sb: #1245-ICSS questions
Fm: Rufus Laggren 76346,1441
To: Bob Cave 73232,2474

 >>  You will get the best accuracy by defining small contexts with words
 >> which are distinct from one another rather than putting all possible
 >> responses into a single context

Yes, does answer my question, point on.  Thanks.  I was wondering what would
improve the "greek" interpretation of "French" over the phone.  Seems by
defining small context for a particular expected response, even 1 sylable
words would be recognized accurately.

Incidentally, how *does* one build a speech model?

Thanks for you quick note.  Rufus

#: 1284 S10/IBM Speech Products ∙5!∙
    08-Mar-94  21:52:44
Sb: #1259-ICSS questions
Fm: Bob Cave 73232,2474
To: Rufus Laggren 76346,1441

Rufus,

>> Incidentally, how *does* one build a speech model?

The truth is, I know very little about that part of the process.  I know that
we start with several thousand recorded utterances and transcripts, and the
whole thing ties up a high-end Risc System/6000 for 24 hours or so.  Beyond
that, it may as well be magic <g>.

Perhaps Brandon knows more than I do on this one?

Regards,

Bob Cave, ICSS Development



#: 1187 S10/IBM Speech Products ∙5!∙
    03-Mar-94  17:05:11
Sb: #1170-ICSS questions
Fm: Stephen Ng 70302,1510
To: Bob Cave 73232,2474

Bob,

Of course what's frustrating/tantalizing about this is that it seems to work
perfectly for a while, then it has periods of not matching at all.  I've
started recording the sessions as you've suggested.

I uploaded two files "Barbe00.wav" and "desse00.wav" which are me saying
"French" and ICSS matching "barbecue" and "dessert".  Can you tell me if there
is any mangling I can do to these files to make them match better?

Thanks,

Steve

#: 1197 S10/IBM Speech Products ∙5!∙
    03-Mar-94  22:29:45
Sb: #1187-ICSS questions
Fm: Bob Cave 73232,2474
To: Stephen Ng 70302,1510

Steve,

I have downloaded the files and will take them to work and listen to them
tomorrow.  It would also be helpful if you could upload your BNF file and tell
me what parameters you are using when you create the context.  This will allow
me to see if I can duplicate the results you are getting.

Thanks,

Bob Cave, ICSS Development

#: 1199 S10/IBM Speech Products ∙5!∙
    04-Mar-94  11:52:07
Sb: #1197-ICSS questions
Fm: Stephen Ng 70302,1510
To: Bob Cave 73232,2474

Bob,

I really appreciate your taking the time to look at this, especially since I'm
really pushing ICSS to its limits.

With the fax I received from Vernon, I hooked up the line in on the Pro Audio
Studio card directly to the Dialogic card this morning.  I *think* I'm getting
cleaner audio; I'll send you a sample file by mail (I can't upload files onto
the conference).

I'll be doing some testing over the next few days to try to get some idea
about whether this improves things.

Yesterday I tried a quick test of the "<cuisine>, please" and "please,
<cuisine>, please" idea but it didn't seem to help things.  Starting a new
conversation for each phone call didn't seem to improve the accuracy much
either.

Thanks again,

--Steve

#: 1244 S10/IBM Speech Products ∙5!∙
    06-Mar-94  20:48:08
Sb: #1199-ICSS questions
Fm: Bob Cave 73232,2474
To: Stephen Ng 70302,1510

Steve,

I got your file and will look at it on Monday.

The other files you sent had a considerable amount of distortion in the
background (an audible hum).  When I brought up the files under the Test
Context function, and looked at the waveform, it was clear that there was a
problem.  The distortion was well above the default thresholds, and ICSS would
have attempted to perform recognition on it.

Regards,

Bob Cave, ICSS Development



#: 1200 S10/IBM Speech Products ∙5!∙
    04-Mar-94  11:52:09
Sb: #1192-ICSS questions
Fm: Stephen Ng 70302,1510
To: Vernon Harvey 72120,2767

Vernon,

Thanks for the fax from the manual page; it's exactly what I needed.  (Makes
me wonder why I paid 30% more for my board to get it direct from Dialogic when
their tech support wasn't that helpful--why so desperate for my 45 bucks?)

I hooked up pin 1 (line 1) and pin 2 (line 2) of P3 to the left and right
signal lines, and pin 20 to the ground line.  Seems to work fine.  But ICSS
appears to respond to only one side...no big deal.

Dialogic told me it's about 3db less than tip and ring coming straight off the
phone line, and that they weren't sure if the 12 volts (or whatever) that you
get when the phone rings would come in through these lines ("probably not, but
find a friend who is handy with a voltmeter").  Well, I tested it myself,
there is no signal on those lines until a connection is made.

I'll be testing today to see if this actually improves my ICSS performance.

Thanks again,

Steve



#: 1207 S10/IBM Speech Products ∙5!∙
    04-Mar-94  16:49:34
Sb: #1200-ICSS questions
Fm: Vernon Harvey 72120,2767
To: Stephen Ng 70302,1510 (X)

Stephen:

If I understand you correctly, it sounds like you are trying to connect a mono
telephone output to a stereo input on your sound card.  The AUD1 is audio
output for telephone line 1, while AUD2 is for your second phone line.
Remember, the Dialogic card supports up to 4 incoming telephone lines on a
single card.  If possible, see if you can connect to the microphone input on
your sound card, as that should be a mono input (Unless you have left and
right mike inputs).

Regards,
Vernon.

#: 1250 S10/IBM Speech Products ∙5!∙
    07-Mar-94  10:25:15
Sb: #1207-ICSS questions
Fm: Stephen Ng 70302,1510
To: Vernon Harvey 72120,2767

Vernon,

My last message wasn't too clear.

I have a phone simulator on line 1 of the dialogic card, and my actual phone
on line 2.  While testing, I want the sound card to take input from line 1,
and during production to take input from line 2.  So I thought I could hook up
line 1 to the left channel, and line 2 to the right.  This works fine as far
as the sound card is concerned, but ICSS doesn't seem to recognize input from
one of the channels (the right channel?).  No big deal, as it just means I
need to plug and unplug phone some phone lines when going from testing to
production.  For real multi-line support, I'll need an audio mixer or an AMX
card.....

--Steve

#: 1256 S10/IBM Speech Products ∙5!∙
    07-Mar-94  21:37:21
Sb: #1250-ICSS questions
Fm: Brandon Booth [IBM] 71042,3144
To: Stephen Ng 70302,1510

Steve

ICSS only deals with mono.  It would be very confused with stereo input.  Or
did I misunderstand your post?

Brandon

ICSS Development

#: 1258 S10/IBM Speech Products ∙5!∙
    07-Mar-94  22:52:23
Sb: #1250-ICSS questions
Fm: Bob Cave 73232,2474
To: Stephen Ng 70302,1510

Steve,

I had a look at the new files you sent today.  They are definitely better
sounding than the previous set, but I did not have much luck using them for
recognition with the BNF you provided (although it worked fine when I used the
microphone).  When I looked at the waveforms under Test Context, there was
still a fair amount of noise showing up.  You may want to consider raising the
Start Talking and End Talking thresholds, since the noise does not seem that
loud.

Regards,

Bob Cave, ICSS Development

#: 1309 S10/IBM Speech Products ∙5!∙
    09-Mar-94  23:31:43
Sb: #1258-ICSS questions
Fm: Stephen Ng 70302,1510
To: Bob Cave 73232,2474

Bob,

Thanks, I will try raising the thresholds and see if that improves things.

Brandon:  Did my subsequent message clear up the stereo input comment I had?
I basically am using the left channel for one phone line, and the right phone
line for another phone line, but I never have both lines active at once
(certainly it would be understandable for this to confuse ICSS).  However ICSS
doesn't seem to hear one of the channels.  This is a minor point for me(it may
even be a glitch with my setup).

--Steve

#: 1 S10/Thread Header
    14-Mar-94  09:50:38
Sb: ICSS Spelling Letters
Fm:
To:

TITLE: ICSS SPELLING LETTERS
KEYWORDS: ALPHABET

#: 1362 S10/IBM Speech Products ∙5!∙
    14-Mar-94  09:50:38
Sb: ICSS Spelling Letters
Fm: Office Automation 100012,3267
To: All


Hello,

I am using ICSS under OS/2 2.1 with a IBM PC 486 66Mhz with 16M of ram.
I need to be able to spell names to ICSS but I have big problems because
the accuracy is very very bad. Is it normal and is there a solution to solve
this problem?

Ian GAMBINI, AMADEUS Company




#: 1366 S10/IBM Speech Products ∙5!∙
    14-Mar-94  14:06:02
Sb: #1362-ICSS Spelling Letters
Fm: Vernon Harvey 72120,2767
To: Office Automation 100012,3267

Ian:

How are you doing the spelling?  Are you using single phoneme based spelling
characters, such as "ay", "bee" "see", etc.  If so, you will probably have
recognition problems as ICSS seems to work best with multi-syllable
recognition (probably because there is more data to recognize).  You may want
to try a military based spelling operation such as "FOXTROT ROMEO ALPHA NANCY
CHARLIE INDIGO EDWARD" to spell "Francis" (I don't think I have the correct
words in the example, but I think you see what I am getting at).  As I recall,
I think there was a dictionary or BNF file in one of the ICSS directories
which contains the police car vocabulary.  That demo used the same sort of
logic to enter license plate numbers, so you could probably scalp the word
list from there.

Regards,
Vernon Harvey.

#: 1379 S10/IBM Speech Products ∙5!∙
    15-Mar-94  04:46:21
Sb: #1366-ICSS Spelling Letters
Fm: Denis Lacroix 100012,3270
To: Vernon Harvey 72120,2767

First, thank you for your answer.
Is there really no other solution to spell letters, because in this case you
have to learn all the correspondance between letters and words ( A-ALPHA,
B-BRAVO,...).
You can't ask everybody to learn this syntax just for spelling.

I am using a DX 66, do you think it will be really better if I use a Pentium ?

Regards,
Ian GAMBINI, AMADEUS Company

#: 1393 S10/IBM Speech Products ∙5!∙
    15-Mar-94  20:52:43
Sb: #1379-ICSS Spelling Letters
Fm: Bob Cave 73232,2474
To: Denis Lacroix 100012,3270

Basically, what Vernon said was correct.  Recognizing individual letters for
the purpose of spelling names is extremely difficult.  There is simply not
enough information available for ICSS to reliably distinguish between the
choices.  The military alphabet was created to solve this problem of speech
recognition between humans. It can be used with ICSS also, and will produce
far better results than saying the letters by themselves.  I recognize the
impracticality of having everyone memorize the military alphabet for the
purpose of talking to a computer.

If there is a very limited set of valid combinations of letters, they can be
added to a context as words (ABC = EY B IY S IY in ICSS phonemes).

Regards,

Bob Cave, ICSS Development



#: 1469 S10/IBM Speech Products ∙5!∙
    19-Mar-94  12:30:04
Sb: #1379-ICSS Spelling Letters
Fm: Rufus Laggren 76346,1441
To: Denis Lacroix 100012,3270

 >> Is there really no other solution to spell letters, because in this
 >> case you

FWI. (I'm no expert).  You have 1000 words to play with for a context (and you
can switch contexts instantly).  So you are not limited to the words used by
police or military.  If you know which letters (or numbers, etc) *must* occur
next, you can limit the selection further, as is suggested elsewhere here.

I don't know if this will help, but you if you favor the letter sound very
much, you may also try *pre*-fixing the actual sound by saying "letter" so the
machine hear "letter-ay", for example.  Not sure this will help or not.

Think how much your understanding of what *you* hear depends on the *context*,
rather than on the sound stream.  You're a lot more complex than ICSS, and I
think you would have trouble gathering data w/out the context.

Best luck, Rufus



#: 1374 S10/IBM Speech Products ∙5!∙
    14-Mar-94  23:28:51
Sb: #1362-ICSS Spelling Letters
Fm: Brandon Booth [IBM] 71042,3144
To: Office Automation 100012,3267

Ian

All speech recognizers that I know about have trouble recognizing single
phoneme utterances.  This is also true of people!  The problem stems from the
fact that there is no data about preceeding and succeeding phones (as it is a
single phone), and so there is no data to distinguish confusible phones (e.g.,
'b' and 'p').  As a subsequent post mentions using the military "foxtrot,
tango" style (or any equivalent) works much better.

Brandon

ICSS Development

#: 1380 S10/IBM Speech Products ∙5!∙
    15-Mar-94  05:01:52
Sb: #1374-ICSS Spelling Letters
Fm: Denis Lacroix 100012,3270
To: Brandon Booth [IBM] 71042,3144


First, thank you for your answer.
Is there really no other solution to spell letters, because in this case you
have to learn all the correspondance between letters and words ( A-ALPHA,
B-BRAVO,...).
You can't ask everybody to learn this syntax just for spelling.

I am using a DX 66, do you think it will be really better if I use a Pentium ?

Regards,
Ian GAMBINI, AMADEUS Company



#: 1398 S10/IBM Speech Products ∙5!∙
    15-Mar-94  23:08:38
Sb: #1380-ICSS Spelling Letters
Fm: Brandon Booth [IBM] 71042,3144
To: Denis Lacroix 100012,3270

Denis

I'm sorry, but a fast machine isn't the answer (at least in this case!).  The
problem stems from the length of the utterance.

Brandon

ICSS Development

#: 1 S10/Thread Header
    24-Mar-94  11:00:06
Sb: ICSS Accuracy
Fm:
To:

TITLE: ICSS ACCURACY
KEYWORDS: CONFIDENCE CERTAINTY RECOGNITION

#: 1534 S10/IBM Speech Products ∙5!∙
    24-Mar-94  11:00:06
Sb: ICSS Accuracy
Fm: Office Automation 100012,3267
To: All 72120,2767


Hello,

I have a big problem with the ICSS recognition, because I have always an
answer. Is there no solution to know if the recognition is good enough to be
accepted. For example, if I say "AVAILABILITY TO TOKYO", sometimes the ICSS
answer is "MOVE UP" !!!  For my project, it's impossible to do something
really good with this problem.

I have also another question. Is it possible to say "I WOULD LIKE AN
AVAILABILITY FROM..." and to ignore all the speech before the word
AVAILABILITY ?
If not, is it possible to ask ICSS which phonemes have been recognized?

Ian GAMBINI, AMADEUS Company



#: 1547 S10/IBM Speech Products ∙5!∙
    24-Mar-94  21:36:39
Sb: #1534-ICSS Accuracy
Fm: Bob Cave 73232,2474
To: Office Automation 100012,3267

Ian,

The current version of ICSS does not return a confidence factor.  It is up to
the application to parse the response to determine if it is usable.  We are
looking into ways to enhance the product in this area.

>>  Is it possible to say "I WOULD LIKE AN AVAILABILITY FROM..." and to
>> ignore all the speech before the word AVAILABILITY ?

No.  ICSS will use all of the speech input for recognition.  It does not
perform word spotting.

>> If not, is it possible to ask ICSS which phonemes have been recognized?

ICSS can not directly return the phonemes that were recognized.  Your best bet
would be to look up the recognized words in the dictionary to get the
associated phonemes.

Regards,

Bob Cave, ICSS Development, IBM

#: 1612 S10/IBM Speech Products ∙5!∙
    28-Mar-94  23:20:50
Sb: #1534-ICSS Accuracy
Fm: Brandon Booth [IBM] 71042,3144
To: Office Automation 100012,3267

Ian

Our recognition is quite good if you have your system configured properly. You
should record a speech file with your mic and sound card setup and listen to
it, looking for excessive noise.  If you continue to have a problem, you may
want to call our tech support line 1-800-627-8363.

Brandon

ICSS Development

#: 1 S10/Thread Header
    30-Mar-94  06:16:29
Sb: Headset
Fm:
To:

TITLE: HEADSET
KEYWORDS:

#: 1640 S10/IBM Speech Products ∙5!∙
    30-Mar-94  06:16:29
Sb: Headset
Fm: Tim M. Heagarty 71541,1432
To: All

Folks,

I am looking for a headset that could be used with the ICSS system and a SB-16
card.  Anyone out there have one that they like?  You could save me a lot of
looking and spec. checking.  I want an operator to be able to sit and speak to
their system and hear the answer without disrupting all others around them.
It should cut down on background noise too.

Thanks for your assistance.

Tim Heagarty

#: 1655 S10/IBM Speech Products ∙5!∙
    30-Mar-94  23:12:20
Sb: #1640-Headset
Fm: Brandon Booth [IBM] 71042,3144
To: Tim M. Heagarty 71541,1432

Tim

The headset that we are familiar with is the Sennheiser 410.  This works quite
well, but is a probably a little expensive.

Brandon

ICSS Development

#: 1 S10/Thread Header
    21-Mar-94  06:47:06
Sb: #1480-ICCS
Fm:
To:

TITLE: ICCS
KEYWORDS: THRESHOLD NOISE NOISY ENVIRONMENT

#: 1482 S10/IBM Speech Products ∙5!∙
    21-Mar-94  06:47:06
Sb: #1480-ICCS
Fm: Ken Solkowitz 72103,2111
To: Richard Frank 73511,3406

Richard -
Do you know how to adjust the thresholds when you're in a noisy environment?
At comdex we use 1500-1750. Here in the office we use around 750. I will check
on the Airborne shipment. The mic was originally for Norm but how about if I
allow you to keep it and use it to promote ICSS the best that you can. Don't
publicize that I gave you the mic. But do publicize if we've been responsive
and helpful. If you need anything else, glossies, fact sheets, demos sent to
someone, please contact me.

Regards,
Ken Solkowitz
IBM Continuous Speech Series
Bethesda, MD 20817
301-803-3897
Fax: [phone]

#: 1486 S10/IBM Speech Products ∙5!∙
    21-Mar-94  11:14:51
Sb: #1482-ICCS
Fm: Richard Frank 73511,3406
To: Ken Solkowitz 72103,2111

 > Do you know how to adjust the thresholds when you're in a noisy
 > environment? At comdex we use 1500-1750. Here in the office we use
 > around 750.

  I went up to 2000 on the thresholds, but no joy.  We used the mixer to
decrease the mic input and increase the speaker output and it worked fairly
well.

 > I will check on the Airborne shipment. The mic was originally for
 > Norm but how about if I allow you to keep it and use it to promote
 > ICSS the best that you can.

  I bought my own mic, but Norm really should have one.  He can reach alot of
people through his show.

 > If you need anything else, glossies, fact sheets, demos sent to
 > someone, please contact me.

  Once I sort through the left-over literature I'll let you know.  If you have
some literature on the IPDS I would like to have it.

  Richard Frank

#: 1 S10/Thread Header
    24-Mar-94  00:00:47
Sb: ICSS not working
Fm:
To:

TITLE: ICSS NOT WORKING
KEYWORDS: ADCIF SOUNDBLASTER SB

#: 1525 S10/IBM Speech Products ∙5!∙
    24-Mar-94  00:00:47
Sb: ICSS not working
Fm: Brian Gorrell 70127,1712
To: Sysop

I just received my copy of the ICSS Developers Toolkit today.  During the test
program I get the following message when I press recognize: A failure was
detected in the ISTG, ISTG restart has been attempted.  I have changed the
configuration parameters to Identification Number "1" and Input Source Name
"icssmmpm".  I loaded the TESTING context.  My configuration is a Pentium
66MHz machine with a Soundblaster 16 SCSI-2 and its default mike (600 ohms
input impedence and sensitivity of 10 to 200mV) running on OS/2 2.1, MMPM/2,
and the service packs for both.  I verified that the mike is working by
recording some speech with the Digital Audio Since I am running the service
pack, I UNPACKed the MEMSH.DLL from the original OS/2 2.1 GA CD.  UNPACK
reported that it did not need to be unpacked so it copied it as is.  The file
size is slightly smaller than the one that came with the CSD.  It looks
correct to me.  What can I do to resolve this?  Thanks.  BTW, I did not find
any mention of this error message in the manual, but since I just got it maybe
I overlooked it.

#: 1546 S10/IBM Speech Products ∙5!∙
    24-Mar-94  21:28:13
Sb: #1525-ICSS not working
Fm: Bob Cave 73232,2474
To: Brian Gorrell 70127,1712

Brian,

The symptom you have is the one that occurs with the CSD version of MEMSH.DLL.
Have you tried recording a file with the ICSS utility for recording files
(under the Speech pull-down)?  I am sitting here at home without my tech
reference, and I cannot remember if MMPM wants the ACDIFNumber for the first
card to be 0 or 1.  Try setting it to 0.

Regards,

Bob Cave, ICSS Development



#: 1551 S10/IBM Speech Products ∙5!∙
    24-Mar-94  23:09:18
Sb: #1546-ICSS not working
Fm: Christopher R. Peck 74214,3567
To: Bob Cave 73232,2474

I received the exact same message (I check the  ACDIF # with the Digital Audio
App under MMPM like the readme said to & it's 1 for my SB16).  I copied the
MEMSH.DLL file from the original OS/2 2.1 CD & still got the same error (I
went out & bought a new mike to be sure it wasn't that).  I then tried the
"Stock Market" & that seemed to work ok (it missed alot of the numbers & said
- but it did pick up what i said).  Any suggestions???




#: 1570 S10/IBM Speech Products ∙5!∙
    26-Mar-94  02:03:01
Sb: #1546-ICSS not working
Fm: Brian Gorrell 70127,1712
To: Bob Cave 73232,2474

Thank you for your response, but I still am having the same problem.  I double
checked the MEMSH.DLL file.  It is 13968 bytes dated 5-07-93 at 12:18pm.  I
copied the Service Pack version to a backup directory before putting the GA
version in the \MMOS2\DLL directory.  The SP version is 14848 bytes dated
2-08-94 at 7:19pm.  So this can not be the problem.  As for the ACDIFNumber,
trying it at 0 made no difference.  I believe 1 is correct though.  I also
tried your other suggestion of recording a file.  The recording went okay and
I then went into the test program and tried to use the recorded file as the
input.  It went through without an error this time, but it recognized "Testing
testing testing, one two three" as "Zero zero six two zero six one two three
six."  I then tried loading the RL context and used the sample WAV file
included and it correctly recognized "Where is the nearest french restaurant."
What else can I do to resolve this?  Thanks.




#: 1608 S10/IBM Speech Products ∙5!∙
    28-Mar-94  21:54:08
Sb: #1570-ICSS not working
Fm: Bob Cave 73232,2474
To: Brian Gorrell 70127,1712

Brian,

Did you record the your sample file at 11 khz ("medium" setting in the record
function, I believe)?  The testing context and the RL context are delivered
for medium bandwidth only.

I will need to check out my setup back at the office to confirm the MEMSH.DLL
sizes, and will post back tomorrow.

Regards,

Bob Cave, ICSS Development, IBM

#: 1635 S10/IBM Speech Products ∙5!∙
    29-Mar-94  22:54:43
Sb: #1608-ICSS not working
Fm: Bob Cave 73232,2474
To: Bob Cave 73232,2474

I installed the OS/2 2.1 Service Pack and the MMPM/2 Service Pack.  The
service pack for MMPM definitely kills ICSS.  I tried using the GA version of
MEMSH.DLL to no avail, ICSS was still broken.  I was able to fix ICSS by using
the SB16 drivers (SB16D2.SYS) from the GA version, however, and they work with
either version of MEMSH.DLL.  Using the CSD version of SB16D2.SYS seems to
cause the problem with ICSS.  The GA version of the driver is dated 5/7/93,
and the size is 41956 bytes.

I have downloaded new drivers from the Creative Labs forum here on CIS, and I
will try those tomorrow (although I think they may be the same as the drivers
that came in the MMPM CSD).

Stay tuned...I should have more info tomorrow evening.

Bob Cave, ICSS Development, IBM

#: 1657 S10/IBM Speech Products ∙5!∙
    30-Mar-94  23:12:26
Sb: #1635-ICSS not working
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Cave 73232,2474

Bob Cave did have good luck with the Sound Blaster drivers he downloaded from
CompuServ.  He'll have to give the details, as I forgot the paper I printed
off at work.

Brandon

ICSS Development

#: 1687 S10/IBM Speech Products ∙5!∙
    01-Apr-94  18:41:54
Sb: #1657-ICSS not working
Fm: Bob Cave 73232,2474
To: Brandon Booth [IBM] 71042,3144

Here is what I found regarding MMPM and ICSS:

For best results, use the GA version of MEMSH.DLL with either the GA
SB16D2.SYS, or the latest version available in the Creative Labs section here
on CIS (sb_os2.exe in library 3, I think).  I was not able to get ICSS to work
with the SP versions of MEMSH.DLL and SB16D2.SYS.

I hope this clears things up.

Bob Cave, ICSS Development, IBM



#: 1706 S10/IBM Speech Products ∙5!∙
    02-Apr-94  13:35:27
Sb: #1687-ICSS not working
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Cave 73232,2474

Thanks for looking into that, Bob.

Brandon

ICSS Development

#: 1614 S10/IBM Speech Products ∙5!∙
    28-Mar-94  23:20:54
Sb: #1546-ICSS not working
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Cave 73232,2474

Bob

For MMPM the ADCIFNumber (Analog to Digital Conversion InterFace Number,
horrible acronym!) of 0 means use the current default, if you have one card
this should be the same as using 1.

Brandon

ICSS Development

#: 1613 S10/IBM Speech Products ∙5!∙
    28-Mar-94  23:20:52
Sb: #1525-ICSS not working
Fm: Brandon Booth [IBM] 71042,3144
To: Brian Gorrell 70127,1712

Brian

The MMPM folks (over on OS2DF1 forum) told us that it was enough to keep the
original memsh.dll, so I'm not sure what the problem is.  You might try using
the GA MMPM.  Do make sure to install the medium models as the supplied
contexts (e.g., the testing context) are compiled in medium.

Brandon

ICSS Development

#: 1712 S10/IBM Speech Products ∙5!∙
    02-Apr-94  16:00:42
Sb: #1613-ICSS not working
Fm: Brian Gorrell 70127,1712
To: Brandon Booth [IBM] 71042,3144

 >> Brian  The MMPM folks (over on OS2DF1 forum) told us that it was enough
 >> to keep the original memsh.dll, so I'm not sure what the problem is.
 >>  You might try using the GA MMPM.  Do make sure to install the medium
 >> models as the supplied contexts (e.g., the testing context) are
 >> compiled in medium.  Brandon  ICSS Development

Brandon

Going back to the GA MMPM fixed my problems, however, I would like to see a
fix for this so that I can use the CSD version of MMPM.  I need it to fix some
SoundBlaster 16 problems with the original GA MMPM.  Only replacing the
MEMSH.DLL as has been suggested will not fix the problem with ICSS.  Thanks.

#: 1748 S10/IBM Speech Products ∙5!∙
    04-Apr-94  21:13:59
Sb: #1712-ICSS not working
Fm: Brandon Booth [IBM] 71042,3144
To: Brian Gorrell 70127,1712

Brian

My Co-worker Bob Cave has had some success using the latest Sound Blaster
drivers available from the Creative Labs section of CompuServ (Go BLASTER I
think), with the corrective service version of MMPM.  You might want to try
that.

Brandon

ICSS Development

There is 1 Reply.

#: 1750 S10/IBM Speech Products ∙5!∙
    04-Apr-94  21:55:22
Sb: #1748-ICSS not working
Fm: Bob Cave 73232,2474
To: Brandon Booth [IBM] 71042,3144

Brandon,

To clarify -- I was not able to get the SP version of MEMSH.DLL to work with
any
SB16 drivers (I know, I told you I did, but when I did further testing, it
turned out to still be broken...worked for two recos, then hung...my
apologies).  I was able to use the SP version of MMPM with the GA MEMSH.DLL,
however, with either the GA SB16 drivers or the new drivers from CIS.

It seems that this is a "your mileage may vary" situation.  Some folks are
able to get things working by only replacing the DLL.  This was not the case
for me, and a couple of other people.  This problem has been reported with
ACPA cards also.  The MMPM folks are working hard to find out what is wrong.

Bob Cave, ICSS Development, IBM

#: 1837 S10/IBM Speech Products ∙5!∙
    09-Apr-94  12:26:58
Sb: #1750-ICSS not working
Fm: Brandon Booth [IBM] 71042,3144
To: Bob Cave 73232,2474

Thanks Bob,

Steve C. is continuing to look at the situation for us.  The problem has
turned out to be more difficult than we first thought, and it is uncertain
when it will be resolved.

Brandon

ICSS Development

#: 1712 S10/IBM Speech Products ∙5!∙
    02-Apr-94  16:00:42
Sb: #1613-ICSS not working
Fm: Brian Gorrell 70127,1712
To: Brandon Booth [IBM] 71042,3144

 >> Brian  The MMPM folks (over on OS2DF1 forum) told us that it was enough
 >> to keep the original memsh.dll, so I'm not sure what the problem is.
 >>  You might try using the GA MMPM.  Do make sure to install the medium
 >> models as the supplied contexts (e.g., the testing context) are
 >> compiled in medium.  Brandon  ICSS Development

Brandon

Going back to the GA MMPM fixed my problems, however, I would like to see a
fix for this so that I can use the CSD version of MMPM.  I need it to fix some
SoundBlaster 16 problems with the original GA MMPM.  Only replacing the
MEMSH.DLL as has been suggested will not fix the problem with ICSS.  Thanks.

#: 1 S10/Thread Header
    11-Feb-94  13:55:33
Sb: Minor ICSS/2 problem.
Fm:
To:

TITLE: MINOR ICSS/2 PROBLEM.
KEYWORDS: CSD

#: 769 S10/IBM Speech Products ∙5 ∙
    11-Feb-94  13:55:33
Sb: Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Brandon Booth 71042,3144

Brandon:

Well, I just received and installed my ICSS for OS/2, however, I seem to have
a minor problem.  I am using a SoundBlaster 16 ASP, with a cheap micro-phone
for now (still waiting for my Sennheiser) on a 80486DX 50MHz VLB system which
has 16M of memory.  The ICSS software installed fine, and the Stock Market
demo works (recognition is a bit weak due to the low quality microphone).

The TSTCTXPM test program does not seem to work.  I can start the program,
load the "TESTING" context, and change the settings to accept input from the
"icssmmpm" device.  When I press the recognize button, and start speaking
immediately, I get a "Spoke too soon" message like I am supposed to get.  If I
wait for the "Start speaking" prompt and then start to speak, no waveform
information information is displayed, and a few seconds after I finish
speaking, I get the message "A failure was detected in the ISTG; ISTG restart
has been attempted".

Any ideas?

Regards,
Vernon Harvey.



#: 774 S10/IBM Speech Products ∙5 ∙
    11-Feb-94  22:30:35
Sb: #769-Minor ICSS/2 problem.
Fm: H. V. Belton 75470,3670
To: Vernon Harvey 72120,2767

Vernon,

Check that :

        ?:\MMOS2\DSP - Where "?" is the Drive where MMPM is installed

was added to your LIBPATH in config.sys



#: 798 S10/IBM Speech Products ∙5 ∙
    15-Feb-94  18:58:14
Sb: #774-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: H. V. Belton 75470,3670

Hello H.V.

I have a "C:\MMOS2\DLL" in the LIBPATH statement, but no "DSP".  I tried
adding "C:\MMOS2\DSP" as you suggested, but still no luck on the TSTCTXPM.EXE.

Any other suggestions are more than welcome.

Regards,
Vernon.



#: 827 S10/IBM Speech Products ∙5 ∙
    16-Feb-94  19:39:50
Sb: #769-Minor ICSS/2 problem.
Fm: Bob Cave 73232,2474
To: Vernon Harvey 72120,2767

Vernon,

We have noticed this symptom when using the latest SB 16 ASP device drivers
provided with MMPM/2 (part of a corrective service package).  Are you using
these drivers?  If so, your best bet may be to go back to the previous level.
We are currently working on a solution.

If you are not using new drivers, post back here and I will see what else I
can find out.

Bob Cave, ICSS Development (sitting in for Brandon on the net for a few days)

#: 862 S10/IBM Speech Products ∙5 ∙
    18-Feb-94  17:35:47
Sb: #827-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Bob Cave 73232,2474

Bob:

Yes, I am using the MMPM/2 CSD.  I will switch back and see it solves the
problem.

Regards,
Vernon.

#: 885 S10/IBM Speech Products ∙5 ∙
    19-Feb-94  21:23:06
Sb: #769-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Hi Vernon

We do have a known problem with the OS/2 2.1 CSD version.  Are you using that
version?  I'm not sure how available this version is.  It seems strange that
the stock demo would work, but not test context as they exercise the same code
path. Do you have PATH and LIBPATH set correctly?  Do you have the environment
variables (ICSSDIR particularly) set correctly.  I kind of grasping at straws
here.  It would be really great if you could write a small demo program and
see how that works.  If you want to register a bug (or for phone tech support)
call our level 1 support at 1-800-627-8363.

Brandon

ICSS Development - IBM

#: 938 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  14:45:43
Sb: #885-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

Bob Cave was kind enough to step in and mention the problem with the CSD for
the MMPM/2 software.  I am currently using the CSD version, so I will revert
back to the original MMPM/2 and see if that resolves the problem.

Regards,
Vernon.



#: 946 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  18:31:56
Sb: #938-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Vernon Harvey 72120,2767

Brandon:

With regards to your suggestion of writing a small demo program, I am also in
the process of doing just that.  I originally tried to compile and use the
RECO.C test program, but dicovered that apparently I must write PM
applications when using ICSS via a SoundBlaster card.  I guess RECO.C is only
valid if using the ACPA card.

If you know of a ways that I can write non-PM applications to work with the SB
card, please let me know, as I am not yet a PM pro.

Regards,
Vernon Harvey.

#: 954 S10/IBM Speech Products ∙5 ∙
    21-Feb-94  20:55:42
Sb: #946-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Vernon

If you are going to use the ICSS MMPM support (as for Sound Blaster), you have
to write a PM application (the decision was forced on us).  Learning PM isn't
really that hard and I recommend giving it a shot.

Brandon

ICSS Development

#: 1125 S10/IBM Speech Products ∙5!∙
    01-Mar-94  16:10:36
Sb: #885-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

Now that the OS/2 2.1 Service Pack is out, I was considering installing it on
my system.  You mentioned that the MMPM/2 CSD had a problem when using ICSS
with a SB16 ASP sound card.  Does this problem still exist with the OS/2 2.1
Service Pack?  I would rather move forward than back to the initial release of
the MMPM/2 software if I can.

Regards,
Vernon Harvey.

#: 1136 S10/IBM Speech Products ∙5!∙
    01-Mar-94  21:47:28
Sb: #1125-Minor ICSS/2 problem.
Fm: Bob Cave 73232,2474
To: Vernon Harvey 72120,2767

Vernon,

We will be checking this out over the next couple of days.  I am not sure if
the problems were reported on a beta of the MMPM/2 CSD or on the released
version.

Bob Cave, ICSS Development

#: 1140 S10/IBM Speech Products ∙5!∙
    01-Mar-94  22:25:07
Sb: #1125-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Vernon

My understanding is that the OS/2 2.1 service pack changes in MMPM don't
coexist well with the current version of ICSS for OS/2.  I'm sorry, but we
don't have any fixes for this problem right now.

Brandon

ICSS Development

#: 1162 S10/IBM Speech Products ∙5!∙
    02-Mar-94  17:09:24
Sb: #1140-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

1:  Is there a fix in the works yet?
2:  Can I still install the Service pack and leave the MMPM/2 at its
    current level (Ie: Initial release before all the problems with
    MMPM/2 CSD and Service Pack software updates?
3:  Is the problem in ICSS or in the MMPM/2 software?
4:  If a fix is in the works, any idea when it will be available?
5:  If I go to an IBM ACPA card, can I use the new MMPM/2?
6:  Should I report this as a problem to IBM support?  Will it help
    to speed up the fix.

I am starting to get into ICSS pretty heavily, but don't want to hit a wall
waiting for the service pack or ICSS fixes if I can get around it. I actually
got my first ICSS PM program working a few nights ago, and am now getting to
really understand the APIs.  It really is pretty slick.

I have noticed a few problems with recognition, but I think it may be related
to a problem with MMPM/2.  I have noticed that when I record audio passages
using the ICSS record or the MMPM/2 Digital audio, I have static in the
recording (It may be in the playback, I'm not sure yet.  I am going to take a
look on the Multimedia forum for OS/2 to see if there is a resolution.

Regards,
Vernon.




#: 1168 S10/IBM Speech Products ∙5!∙
    02-Mar-94  19:16:48
Sb: #1162-Minor ICSS/2 problem.
Fm: Ken Solkowitz 72103,2111
To: Vernon Harvey 72120,2767

Vernon -
I might be over stepping my bounds here and I'll hand off to Brandon as the
"authority" but I think you should call IBM and initiate a problem report.
Basically, they released the MMPM code and it broke existing product code and
potentially your product. That will help.

Regards,
Ken Solkowitz
IBM Continuous Speech Series
Bethesda, MD 20817
301-803-3897



#: 1194 S10/IBM Speech Products ∙5!∙
    03-Mar-94  20:50:31
Sb: #1162-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Vernon

I'm sorry, there is no fix for the problem ICSS and the MMPM service pack
currently available.  I don't know if it makes sense to install the OS/2
update and not install if for MMPM (that is I don't what interaction there
is).  You might ask that question in the OS2DF1 MMPM section.  I don't know if
it is ICSS or MMPM, but we worked well together before MMPM changed, so I (of
course) suspect MMPM :-).  I you use an ACPA card you do not have to use the
icssmmpm input source dll.  You can use the icssacpa input source dll. This
later dll does not use MMPM at all (but is limited to ACPA cards only),
thereby bypassing the problem.  Reporting the problem to to IBM support might
actually help light the right fires (its screwy how things work).

On your static problem, are you using an 8-bit card right now?  That can be a
source of added noise.  Static will definitely interfere with recognition.
Placing the audio card away from other cards and the system fan might help.

Brandon

ICSS Development

#: 1206 S10/IBM Speech Products ∙5!∙
    04-Mar-94  16:32:28
Sb: #1194-Minor ICSS/2 problem.
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

Thanks for your info in the ICSS and MMPM/2 problems.  I guess I will submit a
problem report to IBM to get that ball rolling.  I think in the mean time I
might go buy an ACPA card to ensure that I have something working.

With regards to the SoundBlaster card, it is a 16 bit card in a 16 bit slot,
so I assume that is not the problem.  I will try re-arranging my card to see
if that eliminates the problem with the static.  It is sitting in between my
VLB SCSI and Diamond Stealth card right now, so maybe there is a problem.

Regards,
Vernon.



#: 1229 S10/IBM Speech Products ∙5!∙
    05-Mar-94  14:11:01
Sb: #1206-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Hi Vernon

Moving cards around is one thing I can suggest.  We have seen problems in the
past with placing the audio card near the system fan or other cards which
generate RF (radio frequency) noise.  For instance token ring cards can
generate a substantial hum.  Naturally, we don't know all of the situations
which cause problems.  I should say we have used the Sound Blaster 16 card in
our lab with good results.

Brandon

ICSS Development



#: 1850 S10/IBM Speech Products ∙5!∙
    10-Apr-94  00:42:56
Sb: Minor ICSS/2 problem.
Fm: Bruce Hartwell 72056,1064
To: Brandon Booth [IBM] 71042,3144

I am running into the same error message described in this thread:
"A Failer Has Been Detected In The ISTG, ISTG Restart Has Been Attempted".

I am running ICSS for OS/2 2.1 on an IBM clone 486 33Mhz with 20MB of RAM and
a Soundblaster Pro ( old 8 bit version ).  I have not upgraded to the service
pack yet but
I am using the lastest Soundblaster drivers.

I couldn't tell from reading through the messages if a fix had been found or
not.  Has
a fix or workaround been found? If not, does this problem create any major
side-effects
in applications created using the API?

Thanks in advance for the info.



#: 1855 S10/IBM Speech Products ∙5!∙
    10-Apr-94  18:35:11
Sb: #1850-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Bruce Hartwell 72056,1064

Bruce

The problem that has been described is using ICSS for OS/2 with the OS/2 2.1
CSD version of MMPM/2.  Some people have been able to work around it by using
the original version of MEMSH.DLL.  Others have not gotten this to work.  We
do not have a fix for this problem, and it is currently under investigation.
The problem tends to interfere with recognition.  Is this happening to you?

Brandon

ICSS Development

#: 1856 S10/IBM Speech Products ∙5!∙
    10-Apr-94  21:56:10
Sb: Minor ICSS/2 problem.
Fm: Bruce Hartwell 72056,1064
To: Brandon Booth [IBM] 71042,3144

Brandon:

The sample app that comes with ICSS ( stockmarket ) seems to work (
recognition is
weak, but that is likely due to my low quality board and mic ). I haven't
gotten into
the API enough yet to really test overall performance.

I am in the intial investigation phase of a project that requires fairly
consistent
voice recognition on an OS/2 box.  I was concerned when I read through several
messages describing errors.

Any idea when a fix might be forthcoming?  In your opinion, would this be a
good product to use in developing an educational system that required strong
voice recognition?

Thanks for your help!

#: 1977 S10/IBM Speech Products ∙5!∙
    14-Apr-94  21:27:39
Sb: #1856-Minor ICSS/2 problem.
Fm: Brandon Booth [IBM] 71042,3144
To: Bruce Hartwell 72056,1064

Bruce

Our ICSS for OS/2 works very well on the original OS/2 2.1 with MMPM, if you
use a reasonable card and mic.  A 16 bit card (such as a Creative Labs or
Media Vision 16 bit card) is a better choice than a 8 bit card (because we
have to scale the 8 bit samples to 16; there is more noise in the result and
less information).  We give a list of mics that we've used successfully in the
tech ref.  In evaluating your system, I recommend using the Test Context
program tstctxpm.exe.   It will allow you to interact directly with a context,
and it will display the waveform.  This will allow you to see how much noise
there is in your system.  We also have a record/playback feature.  You should
record and play some files and listen to the results.  If you hear lots of
noise, this will be a problem for speech recognition.

As for the problem with the MMPM CSD, it is being worked on by the MMPM group
and our group together.  I do not have a date for when the problem will be
found (these things sometimes take time).

Brandon

ICSS Development

#: 1 S10/Thread Header
    15-Mar-94  21:00:17
Sb: ICSS & MMPM/2 CSD Fix
Fm:
To:

TITLE: ICSS & MMPM/2 CSD FIX
KEYWORDS: CSD SB16 CREATIVE

#: 1394 S10/IBM Speech Products ∙5!∙
    15-Mar-94  21:00:17
Sb: ICSS & MMPM/2 CSD Fix
Fm: Bob Cave 73232,2474
To: all

Hello all.

Thanks to the folks down in the MMPM development group, we now have a
work-around for the problems you have encountered when using ICSS with the new
MMPM/2 CSD.  Before installing the CSD, copy x:\MMOS2\DLL\MEMSH.DLL to another
directory.  After installing, copy MEMSH.DLL back over the copy included with
the service pack.  According to the MMPM developers, there should be no
negative ramifications of using the  G.A. version of MEMSH.DLL with the CSD.

Regards,

Bob Cave, ICSS Development

#: 1397 S10/IBM Speech Products ∙5!∙
    15-Mar-94  22:56:38
Sb: #1394-ICSS & MMPM/2 CSD Fix
Fm: Steve Weeks 70541,646
To: Bob Cave 73232,2474

 > Before installing the CSD, ........

Thanks Bob, I know about 6 people that will need to know this and will pass it
on.

Steve Weeks

#: 1426 S10/IBM Speech Products ∙5!∙
    16-Mar-94  18:19:55
Sb: #1394-ICSS & MMPM/2 CSD Fix
Fm: AL COLE 71552,2526
To: Bob Cave 73232,2474

I've order ICSS am on back order, but I have already installed the OS/2 2.1
CSD. What should I do when I install ICSS?

Thanks Al Cole

#: 1454 S10/IBM Speech Products ∙5!∙
    17-Mar-94  21:35:04
Sb: #1426-ICSS & MMPM/2 CSD Fix
Fm: Brandon Booth [IBM] 71042,3144
To: AL COLE 71552,2526

Hi Al

If you are using the OS/2 2.1 CSD, you need to save x:\mmos2\dll\memsh.dll
from your GA (pre CSD) installation.  After the CSD install, overwrite the csd
dll with the earlier version.  This information is from the MMPM folks.

Brandon

ICSS Development

#: 1456 S10/IBM Speech Products ∙5!∙
    17-Mar-94  22:31:22
Sb: #1426-ICSS & MMPM/2 CSD Fix
Fm: Bob Cave 73232,2474
To: AL COLE 71552,2526

This refers specifically to the MMPM portion of the CSD.  If you have already
installed it, you should (I think) be able to extract the original GA version
of MEMSH.DLL from your MMPM install disks or CD.  Can anyone confirm this?

Regards,

Bob Cave, IBM
ICSS Development



#: 1544 S10/IBM Speech Products ∙5!∙
    24-Mar-94  21:04:50
Sb: #1456-ICSS & MMPM/2 CSD Fix
Fm: Christopher R. Peck 74214,3567
To: Bob Cave 73232,2474

Perhaps you could place memsh.dll from the GA mmpm disk on here & your ftp
sites
for downloading?  Or simple instructions to get it (do I use unpack?)...



#: 1545 S10/IBM Speech Products ∙5!∙
    24-Mar-94  21:16:45
Sb: #1544-ICSS & MMPM/2 CSD Fix
Fm: Bob Cave 73232,2474
To: Christopher R. Peck 74214,3567

Christopher,

>>Perhaps you could place memsh.dll from the GA mmpm disk on here & your ftp
>>sites for downloading?  Or simple instructions to get it (do I use
unpack?)...

I will look into this...it seems like a good idea.  At the least I will try to
get some instructions for getting the GA version off of the diskettes.  It may
take a couple of days, since I will be out of the office tomorrow.

Bob Cave, ICSS Development

#: 1552 S10/IBM Speech Products ∙5!∙
    24-Mar-94  23:12:24
Sb: #1545-ICSS & MMPM/2 CSD Fix
Fm: Christopher R. Peck 74214,3567
To: Bob Cave 73232,2474

Geez - do you live on here?  I just posted that question...
I have the CD version of OS/2 2.1 & the file didn't appear to be in packed
format on there - but the test program still doesn't work correctly.  The
Stock
program seems to be fine.
Can I get away with the smaller C compiler (the $99 kit) for doing
applications
or do I need the full C/Set & OS/2 Toolkit?
-thanks (I hope)
 Chris

#: 1606 S10/IBM Speech Products ∙5!∙
    28-Mar-94  21:38:58
Sb: #1552-ICSS & MMPM/2 CSD Fix
Fm: Bob Cave 73232,2474
To: Christopher R. Peck 74214,3567

Christopher,

Timing is everything, I guess...

As far as I know, you should be able to use the smaller compiler.  ICSS does
not require the toolkit.

Your problem with the Test Context program is strange.  I will check out one
of the systems at the office and see what I can find out.

Regards,

Bob Cave, ICSS Development, IBM

#: 1610 S10/IBM Speech Products ∙5!∙
    28-Mar-94  23:20:45
Sb: #1552-ICSS & MMPM/2 CSD Fix
Fm: Brandon Booth [IBM] 71042,3144
To: Christopher R. Peck 74214,3567

Christopher

Just a bit of advice, you're going to really wish you had the OS/2 toolkit if
your doing OS/2 programming.  The compiler alone isn't going to cut it.

Brandon

#: 1631 S10/IBM Speech Products ∙5!∙
    29-Mar-94  16:41:30
Sb: #1610-ICSS & MMPM/2 CSD Fix
Fm: Erwin Kiss 72202,666
To: Brandon Booth [IBM] 71042,3144

Brandon,
        I recently bought ICSS Toolkit and I am impressed with its
capabilities. Even though I am only using an 8-bit Sound Blaster Pro card and
a $35 Radio Shack microphone, I still get pretty good results. My question is
about the choice of compilers. I use Borland's C++ for OS/2, do you know if
the ICSS API can be called from Borland? Also, does the IBM OS/2 Toolkit work
with Borland? And finally, how much does the Toolkit cost, and where can I
purchase it?

You have a great product, and I can't wait until I have enough time to speech
enable my simple home-made text adventure games.

#: 1653 S10/IBM Speech Products ∙5!∙
    30-Mar-94  23:01:45
Sb: #1631-ICSS & MMPM/2 CSD Fix
Fm: Brandon Booth [IBM] 71042,3144
To: Erwin Kiss 72202,666

Erwin

I believe you can use the Borland compiler.  I think there are couple of
tricks to know.  I'll ask at work and post the results.  I'm sure the OS/2
toolkit works with the Borland compiler.  I'm sorry, I don't know how much it
costs.  I think you can buy it from Indelible Blue: 1-800-776-8284.

Brandon

ICSS Development

#: 1554 S10/IBM Speech Products ∙5!∙
    24-Mar-94  23:59:14
Sb: #1545-ICSS & MMPM/2 CSD Fix
Fm: Christopher R. Peck 74214,3567
To: Bob Cave 73232,2474

I just install mmpm/2 from my OS/2 2.1 CD on top of the MMPM/2 csd, the test
application works just fine now.  I had tried just replacing mapsh.dll - but -
that didn't fix it (maybe it's the SB16 drivers).  If I have time this weekend
I'll try installing Creative Labs new MMPM drivers & see what happens.

#: 1607 S10/IBM Speech Products ∙5!∙
    28-Mar-94  21:40:35
Sb: #1554-ICSS & MMPM/2 CSD Fix
Fm: Bob Cave 73232,2474
To: Christopher R. Peck 74214,3567

Chris,

I saw this note after posting my reply to your previous note.  Let me know how
it goes with the new drivers from Creative Labs.

Regards,

Bob Cave, ICSS Development, IBM

#: 1611 S10/IBM Speech Products ∙5!∙
    28-Mar-94  23:20:47
Sb: #1554-ICSS & MMPM/2 CSD Fix
Fm: Brandon Booth [IBM] 71042,3144
To: Christopher R. Peck 74214,3567

Christopher

The MMPM folks told us that just using the original memsh.dll with the csd
mmpm would work, so that's what we're telling people.  If it's not working for
you, we need to know.

Brandon

ICSS Development

#: 1793 S10/IBM Speech Products ∙5!∙
    06-Apr-94  17:58:46
Sb: #1611-ICSS & MMPM/2 CSD Fix
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

I just tried the suggested work around for the MMPM/2 CSD, and seem to be
semi-operational.  I can use the record and playback operations on the context
test program, but the test context operation is still giving me the ISTG
failed message.  Any suggestions other than blasting the MMPM/2 and restoring
the original version?

#: 1795 S10/IBM Speech Products ∙5!∙
    06-Apr-94  21:53:15
Sb: #1793-ICSS & MMPM/2 CSD Fix
Fm: Bob Cave 73232,2474
To: Vernon Harvey 72120,2767

Vernon,

Which audio adapter are you using?  If it is the SB16, I suggest also trying
the GA drivers (SB16D2.SYS) or downloading the latest version from the
Creative Labs section (sb_os2.exe in library 3).  I was able to get both of
these drivers to work with ICSS and the GA version of MEMSH.DLL.

Regards,

Bob Cave, ICSS Development, IBM



#: 1806 S10/IBM Speech Products ∙5!∙
    07-Apr-94  13:57:36
Sb: #1795-ICSS & MMPM/2 CSD Fix
Fm: Vernon Harvey 72120,2767
To: Bob Cave 73232,2474

Bob:

Thanks for the suggestion.  Shortly after I posted that last message, I came
read another message from either yourself or Brandon which mentioned just
that.  I have obtained the new SB16 drivers from the Creative Labs Forum, and
will be installing tonight.

Regards,
Vernon Harvey.



#: 1838 S10/IBM Speech Products ∙5!∙
    09-Apr-94  12:27:02
Sb: #1793-ICSS & MMPM/2 CSD Fix
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Vernon

The one thing I can definitely say is that ICSS works very well on the
original OS/2 2.1 with original MMPM/2.  It seems likely that the combination
of the OS/2 2.1 CSD with the original MMPM/2 should work (but is in some ways
not a great solution).  The other possibility of just using the original
MEMSH.DLL with the CSD MMPM has apparently worked in some cases, but not all
(obviously).  I am sorry that is all I know right now.  Detective work on the
problem is going on right now in our lab and in the MMPM lab.  I do not know
when the problem will be solved.

Brandon

ICSS Development

#: 1882 S10/IBM Speech Products ∙5!∙
    11-Apr-94  18:15:58
Sb: #1838-ICSS & MMPM/2 CSD Fix
Fm: Vernon Harvey 72120,2767
To: Brandon Booth [IBM] 71042,3144

Brandon:

I finally got around to installing the Creative Labs SB card updates and
everything now appears to be working properly.

Thanks for your assistance.

Regards,
Vernon Harvey

#: 1979 S10/IBM Speech Products ∙5!∙
    14-Apr-94  21:27:42
Sb: #1882-ICSS & MMPM/2 CSD Fix
Fm: Brandon Booth [IBM] 71042,3144
To: Vernon Harvey 72120,2767

Hi Vernon

I'm glad your up and running.

Brandon

ICSS Development

#: 1 S10/Thread Header
    11-Apr-94  08:21:28
Sb: Big Processing Time
Fm:
To:

TITLE: BIG PROCESSING TIME
KEYWORDS: SPEED GRAMMAR

#: 1859 S10/IBM Speech Products ∙5!∙
    11-Apr-94  08:21:28
Sb: Big Processing Time
Fm: Office Automation 100012,3267
To: All 72120,2767

Hello,

I have a problem with the processing time using ICSS.
I need a quite big grammar and to have good accuracy, I have to compile it
with the High Bandwith Model and with a very small value for the Beam
parameter (0.0000001).
When I say "AVAILABILITY TO NEW-YORK", I have to wait about 3 or 4 seconds for
the response (after saying it).
Is it normal to wait so much ?
I use a 486 DX 66 with 32 Megabytes.

Regards,

Ian GAMBINI, Amadeus Company



#: 1883 S10/IBM Speech Products ∙5!∙
    11-Apr-94  18:22:03
Sb: #1859-Big Processing Time
Fm: Vernon Harvey 72120,2767
To: Office Automation 100012,3267

Ian:

Check your End Silence duration parameter.  This parameter tells the ICSS
software how much silence to wait for before assuming that the user has
stopped speaking.  It probably needs to be a trimmed down a bit.

Regards,
Vernon Harvey

#: 1892 S10/IBM Speech Products ∙5!∙
    12-Apr-94  07:47:52
Sb: #1883-Big Processing Time
Fm: Office Automation 100012,3267
To: Vernon Harvey 72120,2767

Harvey:

First, thank you for your answer, but the End Silence duration parameter was
set to 0.8 seconds (the default value). I tryed to take 0.4 seconds for this
parameter but my problem is the same (less 0.4 seconds). I have to wait about
3 seconds for a small sentence and up to 10 or more if I say something longer.
Is it normal to have to wait so much ?

Regards,
Ian GAMBINI, Amadeus Company

#: 1908 S10/IBM Speech Products ∙5!∙
    12-Apr-94  20:40:28
Sb: #1859-Big Processing Time
Fm: Bob Cave 73232,2474
To: Office Automation 100012,3267

Ian,

>>> I need a quite big grammar

Could you be a little more specific as to how large the grammar is?  The
combination of a large grammar (more choices at any point in the search) and
a high bandwidth context will have a negative impact on how long it takes to
complete a recognition.  Is there any way you can break up your context into
several smaller ones?  This might even allow you to get away with using the
medium bandwidth, which will also help performance.

Regards,

Bob Cave
ICSS Development, IBM

#: 1916 S10/IBM Speech Products ∙5!∙
    13-Apr-94  09:30:32
Sb: #1859-Big Processing Time
Fm: Ora Williamson 71742,2161
To: Office Automation 100012,3267

>>I have a problem with the processing time using ICSS.

Ian: I am from the ICSS development lab.  It sounds like something is up with
your     grammar.  Are you compiling Finite State or Word Pair?  How nested
are your productions?  Would you mind faxing us a copy of your grammar, or
sending us the .BNF and .DIC, so we can test it, tweak it, improve it, and
send it back to you?

The fax number is 301-803-3167.  Send it to my attention.

Regards, Ora Williamson ICSS Technical Liason IBM Spoken Language Systems Lab

#: 1980 S10/IBM Speech Products ∙5!∙
    14-Apr-94  21:58:42
Sb: #1859-Big Processing Time
Fm: Brandon Booth [IBM] 71042,3144
To: Office Automation 100012,3267

Ian

You would be better off with a group of smaller contexts, if possible.  You'll
get better accuracy and better speed.  As has been noted you can play with the
beam width (which is a speed vs. accuracy tradeoff), and the end silence
duration (where you have to be careful of people who speak slowly).  If you
can use medium bandwidth, that is faster.  Also you need to pay attention to
the structure of the bnf.  It is possible to have faster and slower bnf's
which produce the same results.

Brandon

ICSS Development

#: 1 S10/Thread Header
    12-Apr-94  20:45:54
Sb: Installing ICSS from B:
Fm:
To:

TITLE: INSTALLING ICSS FROM B:
KEYWORDS:

#: 1909 S10/IBM Speech Products ∙5!∙
    12-Apr-94  20:45:54
Sb: Installing ICSS from B:
Fm: Bob Cave 73232,2474
To: all

Hi all, I couldn't find the original note on this topic, but I would like to
post the information here anyway...

There was a post last week from a user who ran into problems installing ICSS
for Windows from the B: drive.  There is a bug at the end of the ICSS install,
when it asks for the Win32s diskette.  You can clear this message and complete
the install.  Then, put the Win32s diskette back in the B: drive and invoke
the setup program directly.  The diskette is the standard Win32s distribution
diskette, and has its own setup program.  The ICSS install program simply
invokes it for you as a last step.

Any questions, please post.

Regards,

Bob Cave
ICSS Development, IBM