Developer CD Series 1993…ch: Other People's Memory/ADC Developer CD (1993-03) (''Other People's Memory'')_iso/Dev.CD Mar 93.iso/Development Platforms/CSMP Digests/csmp-v1-048.txt

Text File | 1992-11-18 | 57.5 KB | 1,556 lines | [TEXT/MPS ]

C.S.M.P. Digest             Sun, 12 Apr 92       Volume 1 : Issue 48

Today's Topics:

    TMON 3.0 secret about screen
    Think C 5.0.2 & sprintf
    Whatever Happend to Vol 2: Internet Prog. Guide?
    Big Arrays in Think C
    Radius and AppleTalk PROBLEMS - Anyone else?
    Big Pictures
    OZ & NZ: Pictoids Package 1.0 NOW AVAILABLE for ftp
    Speaker-independent continuous-speech recognition on a Mac !?


The Comp.Sys.Mac.Programmer Digest is moderated by Michael A. Kelly.

These digests are available (by using FTP, account anonymous, your email
address as password) in the pub/mac/csmp-digest directory on ftp.cs.uoregon.
edu.  This is also the home of the comp.sys.mac.programmer Frequently Asked
Questions list.

These digests are also available via email.  Just send a note saying that you
want to be on the digest mailing list to [e-mail], and you will
automatically receive each new digest as it is created.

The articles in these digests are taken directly from comp.sys.mac.programmer.
They are not edited; all articles included in this digest are in their original
posted form.  The only articles that are -not- included in these digests are
those which didn't receive any replies (except those that give information
rather than ask a question).  All replies to each article are concatenated
onto the original article in the order in which they were received.  Article
threads are not added to the digests until the last article added to the
thread is at least one month old (this is to ensure that the thread is dead
before adding it to the digests).

Send administrative mail to [e-mail].

-------------------------------------------------------

From: [e-mail] (Neal Trautman)
Subject: TMON 3.0 secret about screen
Date: 9 Mar 92 15:30:21 GMT
Organization: Farallon Computing, Inc.

>From the MacHack 1991 Hack Contest Stack:
>TMON 3.0 secret about screen
>The TMON 3.0 application loader has a cool secret about screen.
>Be the first to find it and read it, and you will win a prize!
>  Waldemar Horwat

Well, simply goto the about dialog and wait for 3 minutes.
Or you can patch the _Alert filter so you don't have to wait
so long.  This is left as an excercise for the student.  :-)

- -- Neal Trautman
Timbuktu Lead Software Engineer
Farallon Computing, Inc.
[e-mail]

---------------------------

From: [e-mail] (Jack Herrington)
Subject: Think C 5.0.2 & sprintf
Date: 8 Mar 92 02:04:00 GMT
Organization: University Of Miami, Biomedical Computing

I'm almost embarrased to say I have this problem, but I'm using Think C
5.0.2, and I'm tring to sprintf(charTmp,"%g",tempFloat).  And it aint
working, and it isn't the value of the variable.  I've tried setting on
the line above.

I've programmed almost the entire program, minus this stupid little glitcho.

I noticed that ANSI had 'use native floating point' off and my starter
project (in object) had it on, but when I flipped it, still no luck.  In
fact I checked everything, I looked at the code, at all the options, and
I'm mega-stumped.

Any help would be greatly appreciated.
- --
"Electric word 'life', it means forever and that's a might long time.  But I'm
 here to tell yah, there's something else... The after-life, a word of
 never-ending happiness, you can always see the sun, day or night.  So when you
 call up that shrink in Beverly hills, you know the one, Dr. everything-we-all-

+++++++++++++++++++++++++++

From: [e-mail] (Eric W. Douglas)
Date: 8 Mar 92 08:18:34 GMT


[e-mail] (Jack Herrington) writes:

>I'm almost embarrased to say I have this problem, but I'm using Think C
>5.0.2, and I'm tring to sprintf(charTmp,"%g",tempFloat).  And it aint
>working, and it isn't the value of the variable.  I've tried setting on
>the line above.

>I've programmed almost the entire program, minus this stupid little glitcho.

>I noticed that ANSI had 'use native floating point' off and my starter
>project (in object) had it on, but when I flipped it, still no luck.  In
>fact I checked everything, I looked at the code, at all the options, and
>I'm mega-stumped.

I too have had problems with 'sprintf()'... except I was writing code
resources, and getting plain old address errors... I allocated memory,
stepped through machine code... couldn't figure it out...

I seem to remember that there was an update posted on CServe or America
OnLine to the function, but it didn't alleviate my problem.

- --eric

* | Eric W. Douglas             Technojock               [phone] | *
* | I'net: [e-mail]      [e-mail] | *
* | AppleLink: STUDIO.D      Compuserve: 76170,1472       AOL: EWDOUGLAS | *
* |                "if q -> p, and not p, then not q. NOT!"              | *

+++++++++++++++++++++++++++

From: [e-mail] (Roger Allen)
Date: 10 Mar 92 18:57:23 GMT
Organization: Amdahl Corporation, Sunnyvale CA

>I'm almost embarrased to say I have this problem, but I'm using Think C
>5.0.2, and I'm tring to sprintf(charTmp,"%g",tempFloat).  And it aint
>working, and it isn't the value of the variable.  I've tried setting on
>the line above.

try
   sprintf(charTmp,"%lg",tempFloat),
or sprintf(charTmp,"%hg",tempFloat)

I seem to recall having trouble with this before...

Roger
- --
> Roger Allen                   |  All the opinions expressed are my     <
> Amdahl Computer Development   |  own and are not Amdahl's.             <
> [e-mail]           |  ------They paid me to say that------- <

+++++++++++++++++++++++++++

From: [e-mail] (Jack Herrington)
Date: 10 Mar 92 21:54:47 GMT
Organization: University Of Miami, Biomedical Computing

The problem is fixed.  It was that I was having trouble using 'sprint'
with a floating point arguement.  The lesson to be learned is to watch the
option switches when loading in a project, like 'ANSI'.  The standard
'ANSI' uses non-native floating pointer, the 'Starter' OOP project uses
native floating point.  Be forewarned.
- --
"Electric word 'life', it means forever and that's a might long time.  But I'm
 here to tell yah, there's something else... The after-life, a word of
 never-ending happiness, you can always see the sun, day or night.  So when you
 call up that shrink in Beverly hills, you know the one, Dr. everything-we-all-

---------------------------

From: Jim Cook <[e-mail]>
Subject: Whatever Happend to Vol 2: Internet Prog. Guide?
Organization: Prime Computer, Inc.
Date: Mon, 9 Mar 1992 17:23:31 GMT

Whatever happened to the followons to the Internet Programmers Guide?
I remember reading something that said both volume II and volume II were
both in the works due to the amount of material.  I tried sending email
to the author - I believe it was Matthew Xavier Mora - but I haven't
received a reply or a bounced message.  Anyone know anything for real?

Jim
<[e-mail]>


+++++++++++++++++++++++++++

From: [e-mail] (Matt Mora)
Date: 9 Mar 92 19:05:58 GMT
Organization: SRI International, Menlo Park, California

In article <[e-mail]> [e-mail] (Jim Cook) writes:
>Whatever happened to the followons to the Internet Programmers Guide?
>I remember reading something that said both volume II and volume II were
>both in the works due to the amount of material.  I tried sending email
>to the author - I believe it was Matthew Xavier Mora - but I haven't
>received a reply or a bounced message.  Anyone know anything for real?

I'm still here, and yes I am working on it. I haven't been able to spend any
time on it lately because of other projects I am working on. I hope to have
the next version(s) out in time for the WWDC in may.

Sorry for the delay,

Matt



- --
___________________________________________________________
Matthew Mora                |   my Mac  [e-mail]
SRI International           |  my unix  [e-mail]
___________________________________________________________

---------------------------

From: [e-mail] (Sven Grenander)
Subject: Big Arrays in Think C
Date: 9 Mar 92 21:39:35 GMT
Organization: Jet Propulsion Laboratory

There have been a number of suggestions for getting large arrays
in ThC.

What am I missing ? Why not just use Calloc as suggested in
K&R ? You can still index away to your hearts content (array[372][2569]).

Gets a bit messy when you have arrays of arrays of arrays, but no
real problem.

- -Sven

+++++++++++++++++++++++++++

From: [e-mail] (Jerry Frain)
Date: 9 Mar 92 22:50:51 GMT
Organization: Kansas State University

[e-mail] (Sven Grenander) writes:

> There have been a number of suggestions for getting large arrays
> in ThC.

> What am I missing ? Why not just use Calloc as suggested in
> K&R ? You can still index away to your hearts content (array[372][2569]).

> Gets a bit messy when you have arrays of arrays of arrays, but no
> real problem.

Accessing dynamically allocated memory as arrays > 1 dimension can get
messy, since you have to compute the location you wish to reference
when it is normally done manually.

Consider the following two arrays, each of 12 integers:

    int a[3][4], b[2][6];

It should be quite obvious that

    a[2][0]

references the 6th location (starting at 0, of course) of a's space,
since a[2][0] is really a[(2*3)+0] == a[6], if a were linear.

Likewise,

    b[2][0]

references the 4th location of b's space, since b[2][2] is really
b[(2*2)+0] == b[4] if b were linear.

Now, when I allocate the following linear memory space,

    int *c;
    c = (int *) malloc(sizeof(int) * 12);

which location should c[2][0] reference?

The answer is, the C compiler should give you an error because the
dimensions of the space are unknown (actually it's linear, and thus
cannot be calculated for multiple dimensions).

  --Jerry Frain, [e-mail]

+++++++++++++++++++++++++++

From: [e-mail] (Michael J. Wolf)
Date: 10 Mar 92 13:18:04 GMT
Organization: Cardiovascular Image Processing Lab, U of Iowa

In article <[e-mail]> [e-mail] (Jerry Frain) writes:
>[e-mail] (Sven Grenander) writes:
>
>> There have been a number of suggestions for getting large arrays
>> in ThC.
>
>> What am I missing ? Why not just use Calloc as suggested in
>> K&R ? You can still index away to your hearts content (array[372][2569]).
>
>> Gets a bit messy when you have arrays of arrays of arrays, but no
>> real problem.
>
>Accessing dynamically allocated memory as arrays > 1 dimension can get
>messy, since you have to compute the location you wish to reference
>when it is normally done manually.
>

It would noty be messy at all.  You could call calloc for the correct memory
block size and you could use something like the following to do so:

int    *data[x][y]...;

The above would then be a n dimensional array of pointers to ints.  So if you
wanted 3 dimensions you could simple declare:

int    *data[x][y];

Then calloc the memory for the above data.  Then you can index like a 3D array.

With Mac memory manager, you can allocate large handles and declare a type that
is a pointer to an array with the dimensions you wish, then declare a handle
which uses the pointer.

Just one mans opinions

MJW

=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=
 Michael J. Wolf                    |
 Software Engineer                  | McIntosh Jr.:
 Cardiovascular Image Processing Lab|   The Power To Crush The Other Kids
 University of Iowa                 |
 [e-mail]         |Disclaimer?  Hell, I don't even know her!
=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=


+++++++++++++++++++++++++++

From: [e-mail] (Brady Duga)
Date: 10 Mar 92 13:45:51 GMT
Organization: University of Rochester, Rochester NY


One way around this (particularly easy for 2D arrays) would be to store
the length of a row at the head of the array and write a function (or macro)
which returns the correct element. Something like:

int  *a,*b;  //Where 'a' should be a[2][3] and 'b' should be b[3][2]

a=malloc(sizeof(int) * (2*3 + 1));
a[0]=3;
b=malloc(sizeof(int) * (3*2 + 1));
b[0]=2;

Then the function:

GetIndex(arr,x,y)
int *arr,x,y;
{
   return y*arr[0] + x;
}

You could use similar functions to get and set the value of an element. This
makes ther linear array approach much less messy.

- --Brady ([e-mail])

+++++++++++++++++++++++++++

From: [e-mail] (John Bruner)
Organization: CSRD, University of Illinois
Date: 10 Mar 92 09:54:12

In article <[e-mail]> [e-mail] (Jerry Frain) writes:
>
> Consider the following two arrays, each of 12 integers:
>
>     int a[3][4], b[2][6];
>
> It should be quite obvious that
>
>     a[2][0]
>
> references the 6th location (starting at 0, of course) of a's space,
> since a[2][0] is really a[(2*3)+0] == a[6], if a were linear.

Actually (as I'm sure you know) it is the 8th location (starting at 0)

This subject comes up from time to time.

If you want to allocate a large array with fixed dimensions
N1,N2,N3,...,Nn the easiest way is to declare a pointer to an array
with dimensions N2,N3,...Nn.  In this case:

    int (*a)[4];

You can allocate space with

    a = malloc(3 * sizeof *a);

and can refer to elements as a[i][j].

If you want the code to reflect the explicit allocation, you can also
declare a pointer to an array with dimensions N1,N2,N3,...Nn

    int (*a)[3][4];

allocate space with

    a = malloc(sizeof *a);

and refer to elements as (*a)[i][j];
- --
(Dr.) John Bruner, Deputy Director            [e-mail]
Center for Supercomputing Research & Development    [phone] (voice)
University of Illinois at Urbana-Champaign        [phone] (FAX)
305 Talbot Laboratory; 104 South Wright St.; Urbana, IL  61801

+++++++++++++++++++++++++++

From: [e-mail] (Alen Shapiro)
Date: 10 Mar 92 15:25:55 GMT
Organization: Crash TimeSharing, El Cajon, CA

In <[e-mail]> [e-mail] (Jerry Frain) writes:

>[e-mail] (Sven Grenander) writes:

>> There have been a number of suggestions for getting large arrays
>> in ThC.

>> What am I missing ? Why not just use Calloc as suggested in
>> K&R ? You can still index away to your hearts content (array[372][2569]).

>> Gets a bit messy when you have arrays of arrays of arrays, but no
>> real problem.

>Accessing dynamically allocated memory as arrays > 1 dimension can get
>messy, since you have to compute the location you wish to reference
>when it is normally done manually.

...example of a mess deleted...

Try building a real array on the heap...as follows:

#include <stdio.h>
#include <stdlib.h>

void *
ccalloc(elems, sz)
        size_t elems, sz; {
        void *mptr = calloc(elems, sz);

        if(mptr == (void *)NULL) {
                fprintf(stderr, "calloc failed\n");
                exit(1);
        }
        return(mptr);
}

void **
two_d_array(x, y, el_size)
    size_t el_size; {
        int i;
        void **arr = (void **)ccalloc((size_t)x, sizeof(void *));

        for(i=0 ; i < x ; i++)
                arr[i] = (void **)ccalloc((size_t)y, el_size);
        return(arr);
}

main() {
        long **enc, **permenc;

        enc = (long **)two_d_array(4, 4096, sizeof(long));
        permenc = (long **)two_d_array(4, 4096, sizeof(long));

        enc[2][5] = (long)1;
        fprintf(stdout, "%ld %ld %ld\n", enc[2][4], enc[2][5], enc[2][6]);
}

- --alen
[e-mail]

ps you will also have to build a "2_d_free()" but that is easy :-)
pps beware for portability when either <x>*(sizeof(void *)) >= 65536
    or when <y>*el_size >= 65536--compilers with 16-bit ints and
    calloc args of "unsigned" may blow-up above these limits (THINK_C
    does not qualify ... i.e. does NOT blow up).

+++++++++++++++++++++++++++

From: [e-mail] (Stephen Uitti)
Date: 10 Mar 92 20:14:36 GMT
Organization: Interactive Systems, Cambridge, MA 02138-5302

In article <[e-mail]> [e-mail] (Sven Grenander) writes:
>There have been a number of suggestions for getting large arrays
>in ThC.
>What am I missing ? Why not just use Calloc as suggested in
>K&R ? You can still index away to your hearts content (array[372][2569]).

In Think C 3, malloc() took an int for the size of the object,
thus was limited to 32K or 64.  lmalloc() took a long, and could
be used as suggested.  Think Think C 5, malloc() takes a size_t,
which is fine.  This has to do with the evolution of the
language, rather than the compiler.  At the time Think C 3 was
around, malloc() was documented (why?) to take an int.  At that
time, it was generally assumed that int was big enough to hold a
character pointer.  It was true on a VAX, and it was true on a
PDP-11.  I don't happen to know how Think C 4 does it.  NewPtr()
and malloc() are functionally equivalent.  Calloc() clears the
array first, which may or may not be what you want.  Calloc() may
be painful with virtual memory, if initialiation is not required.

Stephen.
[e-mail]

---------------------------

From: [e-mail] (Jim Wintermyre)
Subject: Radius and AppleTalk PROBLEMS - Anyone else?
Date: 10 Mar 92 02:47:04 GMT
Organization: Harvey Mudd College


Hi.  I am working on a program (INIT/driver/DA combo) that uses AppleTalk.
Basically, it allows people to talk to each other over a network in real
time.  The problem I'm having is that on ONE machine I've tried it on,
I can't seem to open the AppleTalk drivers at INIT time correctly.  This
machine is a Mac IIx with 8 megs of RAM and a Radius Two-Page Display.
Some people I've talked to mentioned having experinced similar problems
with AppleTalk and Radius monitors.  Has anyone else had this problem?
Does anyone know what causes it?  Anyone at Radius have any ideas?

Thanks for any help,

Jim Wintermyre
[e-mail]

+++++++++++++++++++++++++++

From: [e-mail] (P. Hawthorne)
Date: 10 Mar 92 08:04:20 GMT
Organization: Reed College, Portland OR


  [e-mail] (Jim Wintermyre) writes:
: The problem I'm having is that on ONE machine I've tried it on,
: I can't seem to open the AppleTalk drivers at INIT time correctly.
: Some people I've talked to mentioned having experinced similar problems
: with AppleTalk and Radius monitors.  Has anyone else had this problem?

  Five will get you ten that this one machine just happens to have lost the
special Radius AppleTalk kludge init that all the others have in their
system folders. If that init is loading and your init continues to fail,
all bets are off and quite possibly, short of a total system overhaul, only
Radius itself could help you out. Best wishes,

  Weighing broad compatibility against delay until entry,
  Theus ([e-mail])

+++++++++++++++++++++++++++

From: [e-mail] (Allen J Michielsen)
Date: 10 Mar 92 21:56:55 GMT
Organization: Syracuse University, Syracuse, NY


- --
Al. Michielsen, Mechanical & Aerospace Engineering, Syracuse University
 InterNet: [e-mail]  [e-mail]
 Bitnet: AMICHIEL@SUNRISE

---------------------------

From: [e-mail] (John Nagle)
Subject: Big Pictures
Date: Tue, 10 Mar 92 19:00:02 GMT
Organization: Netcom - Online Communication Services  (408 241-9760 guest)

     With Color QuickDraw, can picture recording via OpenPicture/
ClosePicture record pictures greater than 32K bytes?  IM V-86
seems to indicate that it can, but never quite says it.

     If you run out of memory while recording a picture under ColorQuickDraw,
what happens?  One hopes they improved on the old approach of simply
overwriting memory.

                    John Nagle

+++++++++++++++++++++++++++

From: [e-mail] (JohnC)
Date: 10 Mar 92 21:05:38 GMT
Organization: The Royal Society for Putting Things on Top of Other Things

In article <[e-mail]> [e-mail] (John Nagle) writes:
>     With Color QuickDraw, can picture recording via OpenPicture/
>ClosePicture record pictures greater than 32K bytes?  IM V-86
>seems to indicate that it can, but never quite says it.

It doesn't need to, since the 32K restriction was lifted with the advent of
the Mac Plus 128K ROMs, documented in IM-IV. ;-)  Check out page 23 of said
volume, where it also talks about how to test if you've run out of memory
while creating a picture.


- --
John Cavallino                  |  EMail: [e-mail]
University of Chicago Hospitals |         [e-mail]
Office of Facilities Management | USMail: 5841 S. Maryland Ave, MC 0953
B0 f++ c+ g+ k s+(+) e+ h- pv   |         Chicago, IL  60637

---------------------------

Subject: OZ & NZ: Pictoids Package 1.0 NOW AVAILABLE for ftp
From: [e-mail] (USENET News System)
Date: Tue, 10 Mar 1992 20:54:46 GMT
Organization: School of Maths & Info. Sci., Massey University, Palmerston North, NZ


Thanks to the help of Hal Miller, OZ & NZ folks don't need to wait for
Pictoids Package 1.0 to work its way through the queues at rascal and
hence onto sumex etc. The package is now avialable for ftp from
shark.mel.dit.csiro.au (144.110.16.11) in pub/Pictoids_1.0.hqx. This
copy will be removed when Pictoids reaches sumex as shark is a sumex
shadow, it should then reappear in shark's sumex space.

Enjoy!

Nigel

- --
- ---
Dr Nigel Perry                    Email: [e-mail]
Department of Computer Science    Tel: [phone] ext 8900
Massey University                 Fax: [phone]
Palmerston North
New Zealand

---------------------------

From: [e-mail] (Mitsuharu Hadeishi)
Subject: Speaker-independent continuous-speech recognition on a Mac !?
Date: 25 Feb 92 17:57:26 GMT
Organization: CERFnet


    The Wall Street Journal yesterday, Monday, 2/24/92,
apparently ran an article about *speaker-independent continuous
speech recognition* running on a MACINTOSH.  It said, according
to what I've heard, that sometime last week John Sculley
demonstrated the technology at some conference.  WHAT THE HELL
IS GOING ON HERE!?  Anyone else have more information on this
development?  This is, naturally, a rather major piece of news,
depending on how true it is.  I'm sure it is not nearly as
impossible-sounding as it sounds, but I'd like to know just
how powerful this technology is (vocabulary size, speed,
etc.)  Who developed this?  Is it based on any existing
speech-recognition theory (Hidden Markov Models, neural-network
based solutions, etc.), or what?

    Apple Link is no help; they just posted a single
paragraph from the article.  Anybody else care to shed some
light on this?



- -------------------------

From: [e-mail] (Roger Allen)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 25 Feb 92 19:19:35 GMT
Organization: Amdahl Corporation, Sunnyvale CA

In article <[e-mail]> [e-mail] (Mitsuharu Hadeishi) writes:
>
>    The Wall Street Journal yesterday, Monday, 2/24/92,
>apparently ran an article about *speaker-independent continuous
>speech recognition* running on a MACINTOSH
...
>etc.)  Who developed this?  Is it based on any existing
>speech-recognition theory (Hidden Markov Models, neural-network
>based solutions, etc.), or what?
>

The article stated that the head developer is Kai-fu Lee who was
hired away from Carnegie-Mellon (sp?) University where he did
research on (you guessed it!) continuous speech voice recognition.
The project at CMU was called Sphinx, I believe, and used custom
hardware.  It seems he has now optimized his algorithms for the '040.

Any comment from the peanut gallery on the CMU Sphinx project?   That
would be the closest thing I would imagine.

Another thing I thought was neat was that you had to address the computer
before giving a command.  Gee! Just like Scotty in Star Trek V!
I hope Paramount doesn't sue Apple because it is infringing on the
"look-and-feel" that it developed.

This sounds just incredible, but then again so did cold fusion.
--
> Roger Allen                   |  All the opinions expressed are my     <
> Amdahl Computer Development   |  own and are not Amdahl's.             <
> [e-mail]           |  ------They paid me to say that------- <



- -------------------------

From: [e-mail] (Bertrand Limoges)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Organization: Sobeco Ernst & Young Inc.
Date: Tue, 25 Feb 1992 19:48:22 GMT

In article <[e-mail]>, Mitsuharu Hadeishi writes:
>
>
>     The Wall Street Journal yesterday, Monday, 2/24/92,
> apparently ran an article about *speaker-independent continuous
> speech recognition* running on a MACINTOSH.
.. [stuff deleted]
> depending on how true it is.  I'm sure it is not nearly as
> impossible-sounding as it sounds, but I'd like to know just
> how powerful this technology is (vocabulary size, speed,
> etc.)  Who developed this?  Is it based on any existing
> speech-recognition theory (Hidden Markov Models, neural-network
> based solutions, etc.), or what?


The info I have is from MacWeek 02.03.92:
1. The project is code-named Casper and runs on a Quadra 900
   souped-up with a digital signal processor (DSP) and a better
   microphone then the standard one shipped with the Macs.
2. Vocabulary is more than 100,000 words.
3. System designed by "sampling hundreds of different voices in a
   wide range of accents, tones and pitches."
4. DSP and mike are likely to be incorporated in new Quadra model
   by year end.
5. Currently requires 1 MB RAM.
6. Problems with "nuances' of speech, such as hesitations...
   Big surprise! 8-)
7. Developped by Apple.

No info on innards 8-(   Hopes this helps, Bertrand Limoges



- -------------------------

From: [e-mail] (Mike Khaw)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 25 Feb 92 19:44:50 GMT

I haven't seen the WSJ article. I started to skim a short article in
the San Jose Mercury (2/25/92) about it, but lost interest when I read
where it said something about speaker-independent discrete word
recognition from a limited/restricted vocabulary.
--
Michael Khaw - Domain=[e-mail], UUCP=...!uunet!parcplace!khaw
ParcPlace Systems, Inc., 1550 Plymouth St., Mountain View, CA 94043



- -------------------------

From: [e-mail] (Norman Graham)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 25 Feb 92 19:37:22 GMT
Organization: Oklahoma State University

In article <[e-mail]> [e-mail] (Mitsuharu Hadeishi) writes:
>
>    The Wall Street Journal yesterday, Monday, 2/24/92,
>apparently ran an article about *speaker-independent continuous
>speech recognition* running on a MACINTOSH.  It said, according
>to what I've heard, that sometime last week John Sculley
>demonstrated the technology at some conference.  WHAT THE HELL
>IS GOING ON HERE!?  Anyone else have more information on this
>development?  This is, naturally, a rather major piece of news,
>depending on how true it is.  I'm sure it is not nearly as
>impossible-sounding as it sounds, but I'd like to know just
>how powerful this technology is (vocabulary size, speed,
>etc.)  Who developed this?  Is it based on any existing
>speech-recognition theory (Hidden Markov Models, neural-network
>based solutions, etc.), or what?

In the Feb 3 issue of MacWeek, there's a front page report of a
preview of this technology given at Demo '92 the prior week. The preview
was given by David Nagel, Apple's Senior VP of the Advanced Technology
Group.

Casper (apparently this technology's code name) was running demo'd
on a Quadra 900 enhanced with a digital signal processor (DSP) and
a better microphone than Apple is now shipping. By year's end, the
DSP and new microphone are expected to be included in a new Quadra
model.

Casper's vocabulary is greater than 100,000 words. It appears that
larger vocabularies need more processing power, so expect a some-
what smaller vocabulary on less powerful machines.

Casper needs about 1 MB of RAM, but Nagle expects to half that
requirement before Casper ships.

No details were given on the implementation technology. I seem to
recall seeing a QuickTime video showing a 3D analysis of speech
that the Casper team is using, but I don't recall what the axis
represented. I also seem to recall reading something that said
the Casper group has taken samples of thousands of people speaking
and they somehow extract essential characteristics of words from
all these samples. This explains why users don't need to train
Casper--the group has already done it.

Disclaimer: I have no knowledge of how speech recognition works,
so don't be surprised if the above is not accurate.
--
Norman Graham

<[e-mail]>                 Standard Disclaimer Applies
{cbosgd,rutgers}!okstate!norman



- -------------------------

From: [e-mail] (Jim Bailey)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Organization: Lexicon, Inc., Waltham, MA
Date: Wed, 26 Feb 1992 05:48:17 GMT

From: [e-mail] (dks=)
Newsgroups: comp.sys.mac.system
Subject: WSJ article on Apple's speech-recognition work
Date: 25 Feb 92 15:06:14 GMT
Organization: Massachusetts Institute of Technology
Nntp-Posting-Host: e40-008-9.mit.edu



 Apple - Unveils 'breakthroughs' in technology
{The Wall Street Journal, 24-Feb-92, p. A3}

> Apple CEO John Sculley said the company had achieved "a
> major breakthrough" by getting its Macintosh computers to
> respond to commands spoken by people using ordinary
> language. Leading computer researchers hailed Apple's
> work, saying it appeared to represent a milestone in the
> decades-long quest to build personal computers that can
> carry out spoken instructions. "As far as I know, this is
> a first," said Marvin Minsky, a computer-science
> professor at the Massachusetts Institute of Technology.
> Mr. Minsky, along with about 500 other people, witnessed
> a demonstration of Apple's technology at the Technology,
> Entertainment and Design conference in Monterey, Calif.,
> Friday evening. About 24 hours earlier, Apple conducted a
> similar demonstration at a Tokyo trade show. Several
> companies, most notably Dragon Systems Inc., already sell
> speech-recognizers for personal computers, but these
> products require special hardware and a speaker must
> "train" these systems to respond to his or her voice.
> Also, these products work well only when speakers say one
> word at a time, pausing after each one. On the other
> hand, Apple's speech-recognizer responded to continuous
> speech, even answering back at times. Mr. Sculley said
> the system works with off-the-shelf Macintosh computers
> without any special hardware such as a
> digital-signal-processor chips and immediately responds
> to a new speaker's voice. It isn't clear how quickly
> Apple can bring its speech-recognizer technology to
> market or whether the company has a significant advantage
> over other rivals. Mr. Sculley described the
> demonstration as a "teaser" that illustrated Apple's
> progress in building what it has called a "knowledge
> navigator," or a pocket-sized computer that can both
> respond to commands and answer questions in ordinary
> language. Mr. Sculley would not comment on the company's
> plans for speech recognition, but he cites Apple's work
> as evidence of continuing innovation in the field of
> personal computing. Observers said that Apple's
> technology should end up in future Macintosh computers or
> in an expected line of "personal electronics," portable
> devices that will serve as electronic notepads or
> electronic references and include some of the functions
> of a cellular telephone. The only potential drawback is
> that the speech processor, at least today, requires a
> top-of-the-line 68040 processor from Motorola, which
> means that any speach-capable computers would sell for
> more than $5,000. "It looks like superb work ... We can't
> do it in our lab," said Wayne Rosing, who heads the
> advanced research lab at Sun Microsystems. Key figure
> Kai-fu Lee, shrugging off compliments, said the
> speech-recognizer could use stand some improvement.
> Untutored speakers might be confused about how to talk to
> their computer without a better interface, "a developed
> model that cues you in on what to say," he said. The
> vocabulary of the system could also be expanded. Mr. Lee
> said that each type of command, such as opening documents
> or inserting words into forms, was associated with
> anywhere from 100 words to 300 words. He did not give the
> size of the system's total vocabulary. Another problem:
> making the speech-recognizer robust enough that
> background noise doesn't throw it off. Apple's
> speech-recognizer requires speakers to begin each command
> by addressing the computer with a prearranged name. Mr.
> Lee's name for his computer helper, or agent, is
> "Casper." At one point in the demonstration, Mr. Lee
> asked Casper to pay two bills electronically, and he
> specified the amount of each check. Mr. Lee's request was
> carried out.


EOF
--
Dhanesh
 <[e-mail]>

--
--
"The Beatles were elevator music in my lifetime 'Yummy Yummy Yummy (I've Got
Love In My Tummy)' had more impact on me" -- Michael Stipe of R.E.M.
[e-mail]



- -------------------------

From: [e-mail] (David Fry)
Subject:  Speaker-independent continuous-speech recogni
Date: 26 Feb 92 19:20:51 GMT
Organization: Harvard Math Department


As I understand it (and I'm no speech expert), Kai-fu Lee's
SPHINX was/is regarded as on, if not "the," leading edge of
speech recognition.  To answer someone else's question, it is
based on hidden markov models.  Several years ago it gave some
impressive results, but none as good as what Apple has
reported now.

For the interested, hidden markov models for speech
are described in the survey article "A tutorial on hidden
markov models and selected applications in speech
recognition" by Lawrence Rabiner in Proceedings of the IEEE
vol.77 no. 2 (Feb. 1989), pages 257-285.  There are references
listed there that describe SPHINX in detail, e.g.
"Large-vocabulary speaker-independent continuous speech
recognition," by K.F. Lee and H.W. Hon, in Conf. Proc. IEEE
Int. Conf. on Acoustics, Speech, and Signal Processing, pp
1205-1208, April 1985.

As a slightly educated observer, I think Apple has a good leg
up on the competition.


David Fry                               [e-mail]
Department of Mathematics               [e-mail]
Harvard University                      ...!harvard!huma1!fry
Cambridge, MA  02138



- -------------------------

From: [e-mail] (David Fry)
Subject:  Speaker-independent continuous-speech recogni
Date: 26 Feb 92 19:24:28 GMT
Organization: Harvard Math Department

>"Large-vocabulary speaker-independent continuous speech
>recognition," by K.F. Lee and H.W. Hon, in Conf. Proc. IEEE
>Int. Conf. on Acoustics, Speech, and Signal Processing, pp
>1205-1208, April 1985.
 ^^^^^^^^^^^^^^^^^^^^^

Oops...that should be pp 123-126, April 1988.


David Fry                               [e-mail]
Department of Mathematics               [e-mail]
Harvard University                      ...!harvard!huma1!fry
Cambridge, MA  02138



- -------------------------

From: [e-mail] (Amanda Walker)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 26 Feb 92 23:29:36 GMT
Organization: Visix Software Inc., Reston, VA

[e-mail] (Mitsuharu Hadeishi) writes:
> I'm sure it is not nearly as
> impossible-sounding as it sounds, but I'd like to know just
> how powerful this technology is (vocabulary size, speed,
> etc.)  Who developed this?  Is it based on any existing
> speech-recognition theory (Hidden Markov Models, neural-network
> based solutions, etc.), or what?

Well, based on what I know of the state of the art, a Quadra 900 with a
DSP chip and a boatload of RAM could do it (in fact, I believe this is
what was used for the demo).  My educated guess would be a fast FFT
running in the DSP, a homomorphic vocoder running on the 040 (or maybe
another DSP), and a large Markov vocabulary model in the 040's RAM, with
the tables compressed as far as they could go and still keep up in real
time.  It would probably take about 16-32 megs of RAM.  (the quote of 1M
that someone else made is ludicrous unless the vocabulary is severely
limited [say, 1K words]).  I also doubt if they break 90% accuracy with
a large vocabulary.

In short, a souped-up Quadra has the raw horsepower, but the
technology's not quite ready to put into an INIT just yet :).

Just for reference, the prevailing view of Apple's speech research that
I've heard from others in the field seems to be "could be neat stuff,
but not at the leading edge of performance yet."


Amanda Walker                              [e-mail]
Visix Software Inc.                    ...!uupsi!visix.com!amanda
--
"Copyright infringement is the sincerest form of flattery."    --Matt Groening



- -------------------------

From: [e-mail] (Donald T. Major)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 27 Feb 92 19:55:56 GMT
Organization: SAS Institute Inc.

All the posts quote Apple as claiming 1 meg of RAM for the product; of
course, the article also seems to claim that a relatively small vocab
was used for the demo (I seem to recall seeing the figure of 2-300
words being mentioned--the article DID say that it wasn't explicitly
made clear by Apple).

--
Donald Major       SAS Institute Inc.  "Cicely, let's fling something!"
[e-mail] SAS Campus Drive                 - Northern Exposure
[phone]     Cary, NC 27513-2414



- -------------------------

From: [e-mail] (David Morgenstern)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 27 Feb 92 21:23:10 GMT
Organization: San Francisco State University

I seem to remember some discussion last year about the other side
of Oskar (?) which is cool voice synthesis. Has anyone seen this
(or heard it)? I thought that the posting said that the work was
based on a different approach to that of other synthesis work...?

Anyone? HAL?

daviD
--
*****     David Morgenstern (a.k.a. BMUG CheerLeader)     *****
*     CIS: 72030,1607   AOL: daviD eM   FAX: [phone]     *



- -------------------------

From: [e-mail] (Evan Torrie)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Organization: CS Department, Stanford University, California, USA
Date: 28 Feb 92 06:33:38 GMT

[e-mail] (David Morgenstern) writes:

>I seem to remember some discussion last year about the other side
>of Oskar (?) which is cool voice synthesis. Has anyone seen this
>(or heard it)? I thought that the posting said that the work was
>based on a different approach to that of other synthesis work...?

  I wish Apple would release this, and get rid of Macintalk for
ever.  I have people who are interested in Berkeley Systems
outSpoken product, but are waiting for a more stable product
than Macintalk.  An Apple rep I know has been promising this
for about the past six months.
   MacWeek had a small piece about it a few weeks ago - it apparently
uses digitised phonemes.

--
- ----------------------------------------------------------------------------
Evan Torrie.  Stanford University, Class of 199?       [e-mail]
Rains RCC - "Remember, even if you win the rat race, you're still a rat."



- -------------------------

From: [e-mail] (Matt Mora)
Subject:  Speaker-independent continuous-speech recogni
Date: 28 Feb 92 17:42:50 GMT
Organization: SRI International, Menlo Park, California

In article <[e-mail]> [e-mail] (David Fry) writes:
>
>As I understand it (and I'm no speech expert), Kai-fu Lee's
>SPHINX was/is regarded as on, if not "the," leading edge of
>speech recognition.  To answer someone else's question, it is
>based on hidden markov models.  Several years ago it gave some
>impressive results, but none as good as what Apple has
>reported now.

Wasn't there a movie on the QuickTime CD that showed the technology
apple was using for Speach recognition? They showed a graphic
view of speach and how certain patterns can be recognized.





Matt



--
___________________________________________________________
Matthew Mora                |   my Mac  [e-mail]
SRI International           |  my unix  [e-mail]
___________________________________________________________



- -------------------------

From: [e-mail] (Larry Rosenstein)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 29 Feb 92 02:50:08 GMT
Organization: Taligent Inc.

In article <[e-mail]>, [e-mail] (Mike Khaw)
writes:
>
> I haven't seen the WSJ article. I started to skim a short article in
> the San Jose Mercury (2/25/92) about it, but lost interest when I read
> where it said something about speaker-independent discrete word
> recognition from a limited/restricted vocabulary.

The following was posted on AppleLink:

Tune in to "Good Morning America" on Monday, from 7:15 to 7:21 a.m. to watch
John Sculley demonstrate Apple's speech recognition technology. John and Kai-Fu
Lee, manager of speech and language technologies, ATG, will show the technology
that has gotten so much coverage in the press lately as a result of their demos
at the TED3 Conference last week in Monterey, and recently in Tokyo.
--
Larry Rosenstein
[e-mail]



- -------------------------

From: [e-mail] (Amanda Walker)
Subject:  Speaker-independent continuous-speech recogni
Organization: Visix Software Inc., Reston, VA
Date: Fri, 28 Feb 92 21:26:53 GMT

[e-mail] (David Fry) writes:
> As I understand it (and I'm no speech expert), Kai-fu Lee's
> SPHINX was/is regarded as on, if not "the," leading edge of
> speech recognition.

The latest word I had was that they were running at about #3, with BBN
and someone else whose name escapes me at the moment playing tag for #1.
I'm sure they had a spiffy demo, but I won't believe "revolutionary"
claims until I see some hard numbers, such as:

How well do they do on something like the NSA standard test data?
Do they break 95% accuracy on 100,000+ word vocabulary?
How much memory does such a vocabulary take, and how much of the CPU
does the recognizer suck down?

Voice recognition that's good enough for navigation and simple task
invocation is much easier that voice recognition that's good enough for,
say, dictation.  The first isn't exciting--the second is.


Amanda Walker                              [e-mail]
Visix Software Inc.                    ...!uupsi!visix.com!amanda
--
"Do not meddle in the affairs of wizards, for it makes them soggy and hard
 to light."    --Robert Anton Wilson



- -------------------------

From: [e-mail] (Matthew T. Russotto)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: Sun, 01 Mar 92 02:06:26 GMT
Organization: University of Maryland, College Park, College of Engineering

In article <[e-mail]> [e-mail] (Larry Rosenstein) writes:
>The following was posted on AppleLink:
>
>Tune in to "Good Morning America" on Monday, from 7:15 to 7:21 a.m. to watch
>John Sculley demonstrate Apple's speech recognition technology. John and Kai-Fu
>Lee, manager of speech and language technologies, ATG, will show the technology
>that has gotten so much coverage in the press lately as a result of their demos
>at the TED3 Conference last week in Monterey, and recently in Tokyo.

Don't take this personally, but "Any sufficiently advanced technology is
indistinguishable from a rigged demo".... (and: caveat emptor)


--
Matthew T. Russotto    [e-mail]    [e-mail]
Some news readers expect "Disclaimer:" here.
Just say NO to police searches and seizures.  Make them use force.
(not responsible for bodily harm resulting from following above advice)



- -------------------------

From: [e-mail] (Arno Gourdol)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 1 Mar 92 10:01:51 GMT
Organization: IMAG Institute, University of Grenoble, France

[e-mail] (David Morgenstern) writes:

>I seem to remember some discussion last year about the other side
>of Oskar (?) which is cool voice synthesis. Has anyone seen this
>(or heard it)? I thought that the posting said that the work was
>based on a different approach to that of other synthesis work...?

>Anyone? HAL?

Oskar? Casper, you mean? Anyway, yes I have heard the MacIntalk
replacement, last year at the WWDC and it was supposed to be made
beta-available during last summer. The quality of the speech was
somewhat better than Macintalk, pefectly understandable, but
yet nottotaly human. Anyway, researchs tend to prove that people
EXPECT computers to speak with a computer voice and that they may
be surprised if their computer really speaks like a human.
The speech synthetiser was also able to do TTS (text to speech)
conversion, ie to read aloud text. Also, you can customize the
voice: ie have a male voice, a female voice, a child voice, or
even different voices per application.
I'm not 100% sure I remember correctly, but the technique used
was to blur two consecutive syllabus sounds to make a new di-sylabus.
To say 'pie' you would blur a digital sampling of 'p' with a
sampling of 'i'. That's a technique that has already been used
by at least a research laboratory in France (France Telecom)
for French. BTW, this technique is localizable, ie it can work
with different languages than English.
The memory requirements were 400 Kb in RAM.

Arno.


--
Arno Gourdol.  <[e-mail]>
Minds, like parachutes, only function when open.



- -------------------------

From: [e-mail] (Keith Rollin)
Subject:  Speaker-independent continuous-speech recognition on a Mac !?
Date: 2 Mar 92 22:10:10 GMT
Organization: Apple Computer Inc., Cupertino, CA

In article <[e-mail]> [e-mail] (Matthew T. Russotto) writes:
>In article <[e-mail]> [e-mail] (Larry Rosenstein) writes:
>>The following was posted on AppleLink:
>>
>>Tune in to "Good Morning America" on Monday, from 7:15 to 7:21 a.m. to watch
>>John Sculley demonstrate Apple's speech recognition technology. John and Kai-Fu
>>Lee, manager of speech and language technologies, ATG, will show the technology
>>that has gotten so much coverage in the press lately as a result of their demos
>>at the TED3 Conference last week in Monterey, and recently in Tokyo.
>
>Don't take this personally, but "Any sufficiently advanced technology is
>indistinguishable from a rigged demo".... (and: caveat emptor)

True, but... As I understand from a conversation at lunch today, the
Macintosh was able to understand voice input from the show's host (Joan
London?) without any training. And to accent the "realness" of the
demo, I understand that the voice synthesis of the Macintosh sucked the
big one (Macintalk quality or worse). If it were a rigged demo, you can
bet they'd be using digitized responses (_I_ wouldn't bet, but _you_
can).

I didn't see the show myself; all of this is second hand.

--
- ----------------------------------------------------------------------------
Keith Rollin           ---            <Taligent .signature under construction>
Disclaimer: Pretty soon, I really _won't_ be speaking for Apple...



- -------------------------

From: [e-mail] (Amanda Walker)
Organization: Visix Software Inc., Reston, VA
Date: Mon, 2 Mar 92 21:39:02 GMT

[e-mail] (Donald T. Major) writes:
> All the posts quote Apple as claiming 1 meg of RAM for the product; of
> course, the article also seems to claim that a relatively small vocab
> was used for the demo (I seem to recall seeing the figure of 2-300
> words being mentioned--the article DID say that it wasn't explicitly
> made clear by Apple).

With that small a vocabulary, it's quite practical, and I'm quite
willing to believe their claims.  This would, however, limit its
usefulness as well.  Even so, it would be quite a studly addition to Mac
system software...


Amanda Walker                              [e-mail]
Visix Software Inc.                    ...!uupsi!visix.com!amanda
- --
It is a vast and wonderful universe, but you wouldn't know it to live here.

- -------------------------

From: [e-mail] (Evan Torrie)
Organization: CS Department, Stanford University, California, USA
Date:  3 Mar 92 06:01:53 GMT

[e-mail] (Keith Rollin) writes:

>True, but... As I understand from a conversation at lunch today, the
>Macintosh was able to understand voice input from the show's host (Joan
>London?) without any training. And to accent the "realness" of the
>demo, I understand that the voice synthesis of the Macintosh sucked the
>big one (Macintalk quality or worse).

  Macintalk quality or worse?!?  I watched the show, and although
the synthesis was stilted, it was 10 times better than Macintalk!
Now, if only Apple would release it (the Text-To-Speech - I can wait
for the voice recognition part).

>--
>------------------------------------------------------------------------------
>Keith Rollin           ---            <Taligent .signature under construction>
                                        ^^^^^^^^^^^^^^^^^^^^
  I thought Taligent officially came into existence today (or was
it last week?).

- --
- ------------------------------------------------------------------------------
Evan Torrie.  Stanford University, Class of 199?       [e-mail]
"I feel slimy already." - John Sununu, on being welcomed by journalists to
his new role at CNN's Crossfire.

- -------------------------

From: [e-mail] (Matthew T. Russotto)
Date: Tue, 03 Mar 92 14:19:24 GMT
Organization: University of Maryland, College Park, College of Engineering

In article <[e-mail]> [e-mail] (Keith Rollin) writes:
>
>True, but... As I understand from a conversation at lunch today, the
>Macintosh was able to understand voice input from the show's host (Joan
>London?) without any training.

Supposedly.  She seemed to be reading from a script, however.  A few times
"Casper" messed up and failed to respond.  But, if I was rigging a demo of
a prototype, I might do that too...

>And to accent the "realness" of the
>demo, I understand that the voice synthesis of the Macintosh sucked the
>big one (Macintalk quality or worse). If it were a rigged demo, you can
>bet they'd be using digitized responses (_I_ wouldn't bet, but _you_
>can).

It wasn't worse than Macintalk quality.  It certainly wasn't digitized, though.
Then again, if I was rigging a demo, I could think of that too...

What it comes down to is that a real breakthrough and a rigged demo are
indistinguishable on TV.  So when does Casper tour the country? :-)

- --
Matthew T. Russotto    [e-mail]    [e-mail]
Some news readers expect "Disclaimer:" here.
Just say NO to police searches and seizures.  Make them use force.
(not responsible for bodily harm resulting from following above advice)

- -------------------------

From: [e-mail] (Amanda Walker)
Organization: Visix Software Inc., Reston, VA
Date: Wed, 4 Mar 92 05:02:51 GMT

[e-mail] (Keith Rollin) writes:
> True, but... As I understand from a conversation at lunch today, the
> Macintosh was able to understand voice input from the show's host (Joan
> London?) without any training.

This isn't surprising.  Most current systems are quite
speaker-independent, although most work a little better with separate
male and female front ends.  It's the "continuous" bit that's a real
bear to get working.  Continuous human speech is much more complex than
it seems at first glance.  'Phonemes' don't really exist in any useful
sense, for example.

[And yes, I do know something about phonetics and phonology--they just
 turn out to be much more useful to humans than to computers :)]

Amanda Walker                              [e-mail]
Visix Software Inc.                    ...!uupsi!visix.com!amanda
- --
UNIX: The only operating system that can be destroyed by mail.

- -------------------------

From: [e-mail] (Alexandre Sasha Polozoff)
Date: Tue, 3 Mar 92 13:03:32 GMT
Organization: Sharpstone Trail Productions - Austin Texas USA


>
> True, but... As I understand from a conversation at lunch today, the
> Macintosh was able to understand voice input from the show's host (Joan
> London?) without any training. And to accent the "realness" of the
> demo, I understand that the voice synthesis of the Macintosh sucked the
> big one (Macintalk quality or worse). If it were a rigged demo, you can
> bet they'd be using digitized responses (_I_ wouldn't bet, but _you_
> can).


The Mac did recognize Joan's voice.  Although it faltered once.

You are right.  The synthesized voice was *awful*.  I would never
be able to handle listening to it.


- --------------------------------------------------------------------------
Alexandre Sasha Polozoff            Internet    [e-mail]
                                    CI$            73467,252
                                    Voice        512/218-1178
- --------------------------------------------------------------------------
Sharpstone Trail Productions        [e-mail]
8903 Sharpstone Trail               uunet!trail.sharpstone.com!info
Austin, TX  78717
- --------------------------------------------------------------------------

- -------------------------

From: [e-mail]
Date: 5 Mar 92 18:41:40 GMT
Organization: Wright State University

In article <[e-mail]>, [e-mail] (Matthew T. Russotto) writes:
...
> Supposedly.  She seemed to be reading from a script, however.  A few times
> "Casper" messed up and failed to respond.  But, if I was rigging a demo of
> a prototype, I might do that too...

Agreed. One element that seemed distinctly _not_ planned happened at the end of
the segment. Caspar seemed to be scanning for its name used in direct address
as a clue to start listening for a command. After the demo Joan and guests were
chatting and Sculley used the name Caspar in a way that (inadvertantly)
triggered the machine. Caspar responded and Joan used the ensuing confusion to
thank everyone and cut to commercial.

Several questions arise:

- - is it Caspar or Casper?
- - can I change the trigger "Caspar" to something else like "Igor" or "Flunky"?

o----------------------------------------------------------------------------o
| John B. Matthews, [e-mail], disclaimer:= myViews <> WSU |
| "I'm a commensal .sig virus, indistinguishable from an ordinary organelle."|
o----------------------------------------------------------------------------o

+++++++++++++++++++++++++++

From: [e-mail] (Jon Pugh)
Date: 6 Mar 92 20:31:45 GMT
Organization: Apple Co.

In article <[e-mail]>, [e-mail] (Keith Rollin) writes:
>
> >Don't take this personally, but "Any sufficiently advanced technology is
> >indistinguishable from a rigged demo".... (and: caveat emptor)
>
> True, but... As I understand from a conversation at lunch today, the
> Macintosh was able to understand voice input from the show's host (Joan
> London?) without any training. And to accent the "realness" of the
> demo, I understand that the voice synthesis of the Macintosh sucked the
> big one (Macintalk quality or worse). If it were a rigged demo, you can
> bet they'd be using digitized responses (_I_ wouldn't bet, but _you_
> can).
>
> I didn't see the show myself; all of this is second hand.

I just went down and got my t-shirt for donating my voice to Casper.  I had
to read a list of 300 phrases into a couple of microphones placed in a
variety of locations.  The phrases were business and Macintosh oriented with
common menu items and meeting oriented phrases predominating.

After it was over, I got to watch a tape of the Good Morning America segment.
It failed a couple of times, so I am inclined to believe it.  There was a
time it wouldn't recognize Joan, so Kai-Fu had to say the command, and there
was another instance where it heard a command that wasn't stated.

All in all, I thought it was a believable demo of pre-release software,
complete with standard demo problems.  Now they just need to make the voice
sound like Majel Barrett.

Jon

+++++++++++++++++++++++++++

From: [e-mail] (Ken McLeod)
Date: 7 Mar 92 09:18:30 GMT
Organization: Apple Computer, Inc.

In article <[e-mail]>, [e-mail] (Jon Pugh) writes:
> I just went down and got my t-shirt for donating my voice to Casper.  I had
> to read a list of 300 phrases into a couple of microphones placed in a
> variety of locations.  The phrases were business and Macintosh oriented with
> common menu items and meeting oriented phrases predominating.

Not just phrases... also grunts, coughs, "umm-hmmm"s, and so forth. :-)

I can verify the "reality" of the Good Morning America demo, having had
the opportunity to [briefly] play with Casper today. It correctly interpreted
my normal conversational voice 100% of the time to perform a bunch of tasks
in the Finder (open and close windows, selecting multiple menu items, and so
on) without having heard my voice before, or requiring me to think about
carefully enunciating my phrases. The smoothness of it blew me away, compared
to my previous experience with using a certain other commercially available
speech recognition device for the Mac (which often ends up "training" the user
to speak loudly and deliberately, and always requires "retraining" for other
users.) Another engineer who was with me managed to duplicate Joan's
experience on the TV demo, but the problem was explained to us and without
going into detail, it seems fairly minor.

Just as as aside, if you saw the demo, you might have gotten the impression
that every phrase you speak will need to be preceded by a control word (i.e.
"Casper".) This isn't the case at all.

>All in all, I thought it was a believable demo of pre-release software,
>complete with standard demo problems.  Now they just need to make the voice
>sound like Majel Barrett.

Agreed. (I did try speaking to Casper in a Patrick Stewart voice. :-)

- -ken

---------------------------

End of C.S.M.P. Digest
**********************