Plugging away with WSR (Windows Speech Recognition)

By George I think I've got it!

Lemme tell ya though, it was not an easy task
groking speech recognition, there are many subtle
ways to fail, as some of you have already found out.

The so-called Help system in WSR is next to useless.

Now that I am well on my way to learning WSR,
I can ask my computer the Really Big Questions
in life, such as:

1) What is the meaning of life.
2) Who invented toilet paper.
3) If all personal computers disappeared,
what real effect would it have on mankind.
4) Is eating beans allowed inside the ISS.
5) Will a USA colony on Mars be democrats or republicans.
6) Now that I can speak directly to my computer by voice,
what should I do if my computer suddenly starts
talking back to me.


I doubt if WSR has any practical uses for the "average user",
but it certainly is fun to play with.

Mark-

--
I was not going to post the following crap, but "andy t"
likes it. It gives him purpose in life, to go into
one of his long abusive tirades; I can't deprive him
of that pleasure.

The following "test" is what I am going to subject WSR to,
about a week from now. To pass the test, no more than
two mistakes can occur, when I "speak in" this article.

It will take me roughly 3 minutes to speak-in this
roughly 200 word medical abstract.

I know, that is a sluggish 67wpm instead of a persons
normal 160wpm normal speaking rate, but this slow
rate is common for intricate technical dictation.

I copied and pasted this article from its original
medical website.

I'm sure most will agree that this is a very stiff
"test" for _any_ speech recognition app.

My bet is that WSR will pass this test, but we
will see for sure in roughly a week from now.

This week is needed by me to get the rest of
my smarts about using WSR.

I know very little about using WSR, as my time has
been spent with other speech apps, namely
Dragon NaturallySpeaking 10.0 and
MacSpeech Dictate 1.5.2



Test abstract begins after line of asterisks
*****************************

Maximizing therapeutic gain with gemcitabine
and fractionated radiation.

Mason KA, Milas L, Hunter NR, Elshaikh M,
Buchmiller L, Kishi K, Hittelman K, Ang KK.

Department of Experimental Radiation Oncology,
The University of Texas M.D. Anderson Cancer
Center, Houston, USA.

mason@notes.mdacc.tmc.edu

PURPOSE/OBJECTIVE: The nucleoside analogue
gemcitabine inhibits cellular repair and
repopulation, induces apoptosis, causes tumor
growth delay, and enhances radiation-induced
growth delay. After single doses of drug and
radiation, maximum enhancement of tumor
response was obtained when gemcitabine
preceded radiation by at least 24 h. Conversely,
the cellular radioresponse of the normal
gastrointestinal epithelium was slightly
protected when gemcitabine and radiation were
separated by 24 h. This differential response
created a time frame within which therapeutic
gain could be maximized. In our present
investigation, we sought to define the most
therapeutically beneficial scheme of gemcitabine
administration when combined with fractionated
radiotherapy.
.... All 3 schedules of drug administration produced
therapeutic gain; however, when gemcitabine was
given more than once in a 5-fraction radiation
treatment schedule, normal tissue toxicity
increased. The highest therapeutic gain (1.4) was
achieved by giving a single dose of gemcitabine
(25 mg/kg) 24 h before the start of fractionated
radiotherapy.

 

"Mark Conrad" <none-of@your-business.invalid> wrote in message
news:310820090509305494%none-of@your-business.invalid...
<!--coloro:blue--><span style="color:blue <!--/coloro-->
> By George I think I've got it!<!--colorc--><!--/colorc-->

Say goodnight to the folks Gracie!

andy t

 

In article <310820090509305494%none-of@your-business.invalid>, Mark
Conrad <none-of@your-business.invalid> wrote:
<!--coloro:blue--><span style="color:blue <!--/coloro-->
> The following "test" is what I am going to
> subject WSR to, about a week from now.<!--colorc--><!--/colorc-->

....plus I posted a lot of blah, blah, blah...

Taking a short break from my frustrating
self instruction, regarding learning WSR.

Todays "school" was for me to create a text-macro
for each and every persons name that appears in
the test article.

These are the names:

Mason KA, Milas L, Hunter NR, Elshaikh M, Buchmiller L,
Kishi K, Hittelman K, Ang KK.


Okay, I created all eight of those macros, they work fine,
100% accurate and very reliable.

For example, I could dictate "Elshaikh M" ten times
and it would come out correct all ten times.

"Elshaikh" is pronounced kinda like el-shake.

I hate using macros, but they are necessary where strange
names are concerned.

Mark-

 

"Mark Conrad" wrote:
<!--coloro:blue--><span style="color:blue <!--/coloro-->
> I doubt if WSR has any practical uses for the "average user",
> but it certainly is fun to play with.<!--colorc--><!--/colorc-->

If you consider yourself an average user, then you're absolutely correct.
However, as I've pointed out many times before, WSR (Windows Speech
Recognition) was designed specifically with the business vocabulary and for
people with disabilities to have easier access to using the computer. I'm not
sure why you ignore these, " facts."

Frankly, the examples of dictation you have posted here over the months have
very little relationship to a typical business user's lexicon. As for your
other posting concerning not liking to create macros for uncommon names, how
else would you propose doing it? In Dragon NaturallySpeaking, when I have to
keyboard in a person's name because it is not known to the lexicon, I simply
highlight it and say, "Make that a shortcut." This just makes it a macro
whether I like to call it a macro or not. I could try to correct the word or
add it to the vocabulary but even in Dragon NaturallySpeaking this is wolf
and painful as it takes several corrections.

Marty Markoe, eMicrophones, Inc.

 

In article <F59107E7-5C2F-41BE-B568-0E78A0F69FF6@microsoft.com>,
<wwwMyMSSpeechcom@discussions.microsoft.com> wrote:
<!--coloro:blue--><span style="color:blue <!--/coloro--><!--coloro:green--><span style="color:green <!--/coloro-->
> > I doubt if WSR has any practical uses for the "average user",
> > but it certainly is fun to play with.<!--colorc--><!--/colorc-->
>
> If you consider yourself an average user, then you're absolutely correct.
> However, as I've pointed out many times before, WSR (Windows Speech
> Recognition) was designed specifically with the business vocabulary and for
> people with disabilities to have easier access to using the computer. I'm not
> sure why you ignore these, " facts."<!--colorc--><!--/colorc-->

What part of my posted phrase "average user"
do you not understand?

<!--coloro:blue--><span style="color:blue <!--/coloro-->
> Frankly, the examples of dictation you have posted here over the months
> have
> very little relationship to a typical business user's lexicon.<!--colorc--><!--/colorc-->

Definition-
typical (adj)
A characteristic example; representative.

You do not consider the medical business a typical business?

If you are placing your own special definition on the word,
such as "small insignificant businesses", let us know upfront.

<!--coloro:blue--><span style="color:blue <!--/coloro-->
> As for your other posting concerning not liking to
> create macros for uncommon names, how
> else would you propose doing it?<!--colorc--><!--/colorc-->

Very easy with Dragon, for example if I wanted to
dictate:

"His name Elshaikh is an unusual name."


I merely say:
"his name literal Elshaikh is an unusual name period"

"literal Elshaikh" is NOT a macro.

It is the spoken form of the new word (new name)
"Elshaikh" added to my custom vocabulary.

A text macro will not work unless the user pauses
both before and after speaking the macro.

No such pausing is needed with my new "word".

Saying: NameLiteralElshaikhVeryUnusual

Yields text: Name Elshaikh very unusual

<!--coloro:blue--><span style="color:blue <!--/coloro-->
> Frankly, the examples of dictation you have posted here over the months
> have
> very little relationship to a typical business user's lexicon.<!--colorc--><!--/colorc-->

People in a typical medical business would disagree.

There are thousands of such medical businesses,
which you are risking antagonizing, because of your
present narrow view of what a "business" is.


<!--coloro:blue--><span style="color:blue <!--/coloro-->
> I could try to correct the word or
> add it to the vocabulary but even in Dragon NaturallySpeaking this is
> wolf
> and painful as it takes several corrections.<!--colorc--><!--/colorc-->

Took me a few seconds, no corrections whatever needed.

 

In article <010920090251155067%none-of@your-business.invalid>, Mark
Conrad <none-of@your-business.invalid> wrote: ...about WSR "test",
an admittedly very tough test.

FINAL TEST RESULTS OF WSR
------------------
(Windows Speech Recognition)


Below is the ORIGINAL test which I dictated
by voice - - - see summary of how WSR
performed near end of this post.


Test is _between_ lines of asterisks
*****************************

Maximizing therapeutic gain with gemcitabine
and fractionated radiation.

Mason KA, Milas L, Hunter NR, Elshaikh M,
Buchmiller L, Kishi K, Hittelman K, Ang KK.

Department of Experimental Radiation Oncology,
The University of Texas M.D. Anderson Cancer
Center, Houston, USA.

mason@notes.mdacc.tmc.edu

PURPOSE/OBJECTIVE: The nucleoside analogue
gemcitabine inhibits cellular repair and
repopulation, induces apoptosis, causes tumor
growth delay, and enhances radiation-induced
growth delay. After single doses of drug and
radiation, maximum enhancement of tumor
response was obtained when gemcitabine
preceded radiation by at least 24 h. Conversely,
the cellular radioresponse of the normal
gastrointestinal epithelium was slightly
protected when gemcitabine and radiation were
separated by 24 h. This differential response
created a time frame within which therapeutic
gain could be maximized. In our present
investigation, we sought to define the most
therapeutically beneficial scheme of gemcitabine
administration when combined with fractionated
radiotherapy.
.... All 3 schedules of drug administration produced
therapeutic gain; however, when gemcitabine was
given more than once in a 5-fraction radiation
treatment schedule, normal tissue toxicity
increased. The highest therapeutic gain (1.4) was
achieved by giving a single dose of gemcitabine
(25 mg/kg) 24 h before the start of fractionated
radiotherapy.

*******************
End of test - - -




SUMMARY OF TEST
************

Speed - Very low, about 33wpm, consistent with
what "andy t" mentioned in one of his posts.
(he posted 20wpm to 30wpm if I recall correctly)

Raw accuracy before any correction, very poor.
Eleven mistakes total in the roughly 200 words,
yielding raw accuracy of about 94.5%.

The mistakes from top to bottom in the flawed
version (below) were as follows -
-----------------------
game S/B gain (line 1)
used an S/B Houston (line 7)
analogs S/B analog (line 9)
after a single doses S/B after single doses (line 13)
game S/B gain (line 22)
shot S/B sought (line 23)
.... All S/B ... All (line 27)
three S/B 3 (line 27)
game S/B gain (line 28)
treatments S/B treatment (line 30)
game S/B gain (line 32)





(flawed version below)
***************

Maximizing therapeutic game with gemcitabine (line 1)
and fractionated radiation.



Mason KA, Milas L, Hunter NR, Elshaikh M,
Buchmiller L, Kishi K, Hittelman K, Ang KK.

Department Of Experimental Radiation Oncology,
The University Of Texas M.D. Anderson Cancer
Center, used an, USA. (line 7)

mason@notes.mdacc.tmc.edu

PURPOSE/OBJECTIVE: the nucleoside analogs (line 9)
gemcitabine inhibits cellular repair and
repopulation, induces apoptosis, causes tumor
growth delay, and enhances radiation-induced
growth delay. After a single doses of drug and (line 13)
radiation, maximum enhancement of tumor
response was obtained when gemcitabine
preceded radiation by at least 24 h. Conversely,
the cellular radioresponse of the normal
gastrointestinal epithelium was slightly
protected when gemcitabine and radiation were
separated by a 24 h. This differential response
created a time frame within which therapeutic
game could be maximized. In our present (line 22)
investigation, we shot to define the most (line 23)
therapeutically beneficial scheme of gemcitabine
administration when combined with fractionated
radiotherapy.

.... All three schedules of drug administration (line 27)
produce therapeutic game; however, when (line 28)
gemcitabine was given more than once in a
five-fraction radiation treatments schedule, (line 30)
normal tissue toxicity increased. The highest
therapeutic game (1.4) was achieved by giving a (line 32)
single dose of gemcitabine (25 mg/kg) 24 h before
the start of fractionated radiotherapy.


SUMMARY OF SUMMARY
***************

People, you will have to take my word for it, dragon would
have yielded a speed on this severe test of about 67wpm.

The raw accuracy would have been 99%.
(2 mistakes for this 200 word article)

Some here will say that this was too severe a test to impose
on WSR, however I wanted to see just how good WSR is
as a business alternative to the very high priced medical
Dragon which costs a minimum of $1,600

I agree it was too severe a test.

WSR will perform much better on everyday words.


All in all, I am very impressed with WSR.

If any here doubt the validity of my results, by all means
run your own test.

The conditions I adhered to were that I would NOT subject
WSR to the test article ahead of time.

Adding and training individual words is okay.

Making individual very short macros is okay, for things like
proper names, or specific websites.

I used 8 macros for 8 peoples names.

I used one macro to print: mason@notes.mdacc.tmc.edu

I used one macro to get "hours" to print as "h".

I used one macro to print "mg/kg"

Eleven macros total.

Mark-

 

In article <020920090712225015%none-of@your-business.invalid>, Mark
Conrad <none-of@your-business.invalid> wrote:
<!--coloro:blue--><span style="color:blue <!--/coloro-->
> FINAL TEST RESULTS OF WSR
> ------------------
> (Windows Speech Recognition)<!--colorc--><!--/colorc-->


REVISED final results of WSR "test"

Okay people, I am going to have to eat crow,
in order to give WSR the benefit of the doubt.

"andy t", pay attention !!!

Several mistakes I made during that "test".

If I had the smarts to say "numeral three"
then that would have eliminated one mistake
where my dictation required "3" instead
of "three".

The word "gain" would have been properly
recognized if I would have spoken it as a
hard "a" as in "gaaan", by grimacing as
in a smile. That would have eliminated
FOUR mistakes.

Total mistakes would have dropped to 6,
yielding a respectable 97% raw accuracy
for the 200 words of that severe test.

I was off on the average speed also, as the
entire dictation could be done in 4 minutes,
yielding a respectable speed of 50wpm
for that complicated medical test.

Lemme say that dictating that complex medical
example in 4 minutes will make a blubbering
wreck out of anyone who attempts it.

Even high-priced Dragon can't do much better than
about 67wpm on highly complex medical dictation.
(for me, at least)

....well, really pushing Dragon I might get that
complex medical dictation done in 4 minutes
yielding 100wpm, but again I would need to go
to a speech therapy session afterwards, to get
the knots in my tongue untied.

One WSR quirk I found, if others want to duplicate
this test on their own computers:

WSR wants to print "radio response" while this
dictation demanded "radioresponse".

I found out the hard way that the NoSpace command
will not work unless I spoke it this way:

"radio.....no.....space.....response"

In other words, for me radioNoSpaceresponse
spoken at a normal fast rate failed 100%
of the time.

Mark-

 

In article <4a9f103c$1@news.x-privat.org>, Mr.Pancks
<pancks@bleedingheart.yard> wrote:
<!--coloro:blue--><span style="color:blue <!--/coloro-->
> What the fudge are you doing?<!--colorc--><!--/colorc-->

Having fun, mainly. Learning how to grok WSR
as a side benefit of that fun.

Until recently, thought WSR was a steaming POS.

Now realize it is good enough for trivial hobby use.


<!--coloro:blue--><span style="color:blue <!--/coloro-->
> You have a speech to text program that you have to
> constantly test by talking to yourself?<!--colorc--><!--/colorc-->

Nope, talk to the speech app, which is a bit different.

Not above talking to myself though, because it is nice
to hear someone intelligent speaking.

<!--coloro:blue--><span style="color:blue <!--/coloro-->
> Get a cat. <!--colorc--><!--/colorc-->

Had one, darn thing never said a word.



So how come you don't fool around with useless toys
like Windows Speech Recognition. (WSR)

No sense of adventure?

Mark-

 

=?Utf-8?B?d3d3Lk15TVNTcGVlY2guY29t?=
<wwwMyMSSpeechcom@discussions.microsoft.com> wrote:
<!--coloro:blue--><span style="color:blue <!--/coloro-->
> "Mark Conrad" wrote:
> <!--coloro:green--><span style="color:green <!--/coloro-->
>> I doubt if WSR has any practical uses for the "average user",
>> but it certainly is fun to play with. <!--colorc--><!--/colorc-->
>
> If you consider yourself an average user, then you're absolutely
> correct. However, as I've pointed out many times before, WSR
> (Windows Speech Recognition) was designed specifically with the
> business vocabulary and for people with disabilities to have
> easier access to using the computer. I'm not sure why you ignore
> these, " facts." <!--colorc--><!--/colorc-->

Maybe because they are not facts, Spammer. Windows Speech
Recognition was designed specifically for "casual users". That
fact is straight from the lead Microsoft programmer.