Vista - Re: Decrease input level of a USB connected microphone

"Rick Rogers" wrote:
I believe Mark is using one of the standard USB sound cards made
specifically for speech recognition. Let him correct me if I'm wrong but it
is likely either one of the Andrea, Buddy, or VXI USB pods. These USB sound
pods are designed specifically for the frequency range of the human voice and
work very well for speech recognition. The best thing about these USB sound
pods is they use the generic USB audio sound driver built into Windows (XP
and Vista). When these USB pods were used in XP, they did not have the
problem where the audio level was reset unknowingly. The audio level in Vista
remains correctly set when using Dragon NaturallySpeaking.

I was invited out to Redmond in the summer of 2006 to preview Windows speech
recognition being built into the Vista operating system. At that time and
currently, we have noticed the problem with the audio subsystem setting the
microphone to a level much too high for WSR.

Although it is quite annoying, we run the audio setup in WSR (Windows®
Speech Recognition) every time we use it. This is the part where you, "Set up
microphone," and read Peter dictates to his computer.....

On a microphone/USB pod combination we manually open the microphone level
and set it at one half whatever Windows sets it to. In the case of
theBoom/Andrea USB-SA sound pod found at:
http://www.mymsspeech.com/microphone...asp?prodID=146
Windows sets it at the maximum 100. We manually set it to 50 and it still
works great for accuracy but more importantly it does not automatically turn
a microphone back on when we put it to sleep and it does not automatically
page up, page down or any of the other weird things it is known to do when a
microphone level is too high.

You can set the microphone level manuall by:
1. Right click on the speaker icon in the lower right Taskbar
2. Choose Recording Devices
3. Double click the microphone device you are using (USB or soundcard)
4. Click the levels tab and set that to one half what it was after running
Set up microphone.

Marty Markoe, eMicrophones, Inc.
http://www.mymsspeech.com

In article <A0259AD5-BDDE-46F7-BB6E-25CE7F447ECC@xxxxxx>, Marty
Markoe - eMicrophones, Inc.
<MartyMarkoeeMicrophonesInc@xxxxxx> wrote:
Yes, a VXI Translator from "VXI Corporation Parrott U"

This same external USB sound card is used with Dragon 10.0
full medical software, which does not suffer at all from
stray background noises.

Same external sound card is also used on the Mac OS side
with "MacSpeech Dictate 1.2.1" which also does not suffer
from the stray background noise problem that afflicts WSR.


WOW !!! - - - just now I turned my TV on, softly in the
background, heh heh...

Vista Speech is carrying on quite a conversation with itself,
from that modest background noise.

Here is some of the text that is being generated:

"with us family's years in the SMS is a few examined and who
will have some who could have all you have these are as is
and has a has is is he who you are the two has a you has our"

....etc., etc., etc.

I checked the level setting afterwards in your step-4, it was still
at "zero" where I had set it.

This "zero" setting is not remembered, BTW, if Vista Speech is
shut down and then re-opened, the level has to be reset again.

The only way I know the zero level setting is having _some_ effect
is that the mic level in WSR while speaking is definitely lower
according to the blue vertical speech bar just to the right of
the microphone icon in Vista Speech - - - the bar that shows the
strength of my voice while speaking into Vista Speech.

Mark-

"Mark Conrad" wrote:

You should not have to set the volume slider down to zero. Try running the,
"Set up microphone," with the level slider open and showing. And after you
click Finish, said the slider to ½ the amount. If it is set to 80, set it to
about 40%.

My professional office building has quite a noisy blower system for the
heating/air conditioning. You would think the problem would be that it is
cycling on and off all day. However with this newly found technique of
cutting the volume level in half, it does not seem to matter if the blower is
on or off.

Have you tried another microphone with WSR?

Marty

In article <ED6E937A-A4B5-4B18-9116-8C514BF995DA@xxxxxx>, Marty
Markoe - eMicrophones, Inc.
<MartyMarkoeeMicrophonesInc@xxxxxx> wrote:
Naw, that won't work Marty, it boils down to being the same
volume slider that I was setting to zero manually

Let me explain. If a user runs "set up microphone" then shouts
in the test sentence "Peter dictates to his computer..." etc.,
then WSR will automatically set the mic level about "10" or
even less, depending on how loud the user is shouting.

Likewise, if the user speaks in the test sentence very softly.
he can get WSR to automatically set that same slider level
to 100%, all the way to the right.

Yes, results are the same, WSR itself is mishandles background
noise, and/or WSR refuses to let the user manually set the
level of the microphone low enough to allow me to place the mic
extremely close to my mouth (one quarter of an inch)
in order for me to weaken background noises to the maximum,
compared to my voice.

In other words, when using the same mic, the same technique in
setting up microphone, the same OS (Vista Ultimate), the same
computer, WSR persists in changing background noise to text,
while Dragon does not.

I think I may be confusing you Marty. All I am saying is that
WSR can not properly handle the same level of background noise
that Dragon can, everything else being equal.

Of course I can get WSR to perform flawlessly in a noisy
environment, same as you do, however if I increase the
background noise beyond a certain point, then WSR turns
to poo poo, while Dragon handles the increased noise nicely.

Microsoft needs to work on whatever part of the WSR code
that is causing WSR to turn obvious noise into text.

WSR code is too "loose", it will output a text word even when
the noise sounds NOTHING LIKE the text it is outputting.

That is unacceptable, weakens WSR's value in noisy environments.

Of course, even with Dragon, if a user really cranks up background
noise beyond a certain point, then even Dragon will suffer the
consequences, however WSR remains just too darn "bad" at
properly handling background noises.

Mark-

"Mark Conrad" wrote:
After further testing yesterday I have to agree you are absolutely correct.
Background noise in my office usually never gets worse than what I would call
moderate levels. This is when our shipping manager about 20 feet away starts
packaging orders to go out. The loudest noises actually when he rips the
packing tape from the tape gun. Up to that level I have no problem.

Yesterday I played a recording made in a noisy brokerage stock trading
office. I had the noise coming from left and right at about 95 dB. Even with
three different highly noise canceling microphones, and with the volume
sliders set to zero (they magically set themselves back up to 100 somehow)
WSR became much less accurate.

I have to visit a demonstration by a major manufacturer today in New York
City, but tomorrow I will write up a report for submission to the speech
recognition workgroup at Microsoft. They are very open with me about issues
they are allowed to talk about and I look forward to hearing back from them
and will report here if I'm able to.

Thanks for bringing this issue to the forefront.

Marty

In article <EB8B5DCF-489C-4E8B-A804-03484727D318@xxxxxx>, Marty
Markoe - eMicrophones, Inc.
<MartyMarkoeeMicrophonesInc@xxxxxx> wrote:
Thanks very much for that effort Marty, I would like to see
Vista speech succeed, if for no other reason than to put
pressure on Nuance.

Mark-

"Mark Conrad" wrote:
A prominent member of the speech recognition workgroup responded quickly to
our concerns. They wrote:

"1. Microphone volume can be changed by AGC or by another applications. You
can turn AGC off and play with it to see if it chances anything, if your
microphone has a “boost� feature, turn it off too. If you use same microphone
for other applications, consider to use a dedicated microphone for speech."

I responded that using an Andrea USB-SA full duplex pod does not have an AGC
setting or boost. Also, the microphone is only used for WSR.

"2. I have read a few of your posts, you had good tips thereJ I would like
to add one. When microphone wizard in Vista doesn’t work well for my
microphone, I open GoldWave to record something and ensure the volume level
is at about 30%. This has been working pretty well for me."

I downloaded GoldWave yesterday and will try this later today.

Before leaving the office yesterday I found in the Advanced tab of
Microphone Properties, settings Exclusive Mode were enabled for:
1. Allow applications to take exclusive control of this device
2.Give exclusive mode application priority

Have you played with these? I will try disabling/enabling the various
combinations and see if that helps.

Marty Markoe, eMicrophones, Inc.
See us at: http://www.mymsspeech.com

PS. It seems this thread would be more appropriate at:
microsoft.public.speech.desktop
so I'll post over there as it may be easier to find the posts and respond to
them.

In article <88210970-A408-4F5D-AA5A-1B02BA5A4708@xxxxxx>, Marty
Markoe - eMicrophones, Inc.
<MartyMarkoeeMicrophonesInc@xxxxxx> wrote:
Good luck Marty, try setting GoldWave all the way down
to zero to see if it can bypass the regular Vista mic level
controls; that would be good enough for our purposes.

If that works, we can creep upwards on the level control
and possibly improve matters by rendering the mic level
so weak that it responds to our voice but ignores background
noises a bit better.


No, I chickened out on those settings, guess I should have tried them.


Sounds like the Microsoft recognition workgroup guys
did not see fit to address the root cause of the problem
in public - - - the root cause being WSRs tendency to
output text when the background noise does not sound
anything like the text that is being output.

Hopefully they will address that point in private,
amonst themselves.

Mark-