About spectrum of speech and fax transmission

hi all,
Someone please tell me why the spectrum of speech rolls off
towards the higher frequencies ?????? I am a newbie to DSP so, 
simple terms please as far as possible.

I am hoping to use this fact to distinguish it from other
data transmissions such as fax. 
(this was suggested to me by Mr.Vladimir Vassilevsky on Comp.DSP only)

Yes Mr. John I am thankful to u also for ur reply. I am trying to read 
about ITU-T T.30 protocol but could not find a good enough source online.

Mr.John had suggested using the two tone (frequency) handshake between two
fax machines as a way of identifying fax transmissions. 

All ideas are welcome and appreciated.

Hi,
if you are looking for info how fax technology is different from voice
modem communication the answer is simple: they used different signals
what controlled by correspondent international standards. Read the
standards and you will know everything about voice and fax
transmission. You don't need to invent a wheel in the topic.
Good luck,
Alex

>Hi,
>if you are looking for info how fax technology is different from voice
>modem communication the answer is simple: they used different signals
>what controlled by correspondent international standards. Read the
>standards and you will know everything about voice and fax
>transmission. You don't need to invent a wheel in the topic.
>Good luck,
>Alex
>
>
Hi,

Please tell me if there is some online resource where I can find
these international standards, because I couldnot find the actual
ITU-T T.30 fax transmission protocol at all - only annoying
referals to it being conformed to.
 Thx in advance.

Hi,
T.30 is very old protocol and, you are right, I was not able to google
one, too. The only way to get the text happened with me when I used to
work as a modem developer at Cirrus Logic.
To get voice protocols is easy job. Look on ETSI site. They have
everything, including homologation, at least 5 years back, when I
maintained Intel modems I downloaded about 600 pages of ETSI standards
covering all modem protocols except fax. :(
Good luck,
Alex

VSP wrote:

> hi all,
> Someone please tell me why the spectrum of speech rolls off
> towards the higher frequencies ?????? I am a newbie to DSP so, 
> simple terms please as far as possible.

This is the scientific fact. Throat is made of soft materials.

> 
> I am hoping to use this fact to distinguish it from other
> data transmissions such as fax. 

Sure. This simple method works very well.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Why don't you make some recordings and have a look at them with
something like CoolEdit.  You will very quickly see the distinctive
features of speech versus fax or modems, especially if you use the
spectrum view in CoolEdit or whatever you use.

John

VSP wrote:
> hi all,
> Someone please tell me why the spectrum of speech rolls off
> towards the higher frequencies ?????? I am a newbie to DSP so, 
> simple terms please as far as possible.
> 
> I am hoping to use this fact to distinguish it from other
> data transmissions such as fax. 
> (this was suggested to me by Mr.Vladimir Vassilevsky on Comp.DSP only)
> 
> Yes Mr. John I am thankful to u also for ur reply. I am trying to read 
> about ITU-T T.30 protocol but could not find a good enough source online.
> 
> Mr.John had suggested using the two tone (frequency) handshake between two
> fax machines as a way of identifying fax transmissions. 
> 
> All ideas are welcome and appreciated.

You seem to be making this really hard by not explaining what you are 
really trying to achieve.

If you hear the whole call, you can distinguish FAX by the pulsing of 
1100Hz from the caller (almost always present), or by a burst of 2100Hz 
from the answerer (often suppressed).

Do you need to distinguish quickly at the start of the call, so you can 
select how the call is handled? If so, the above tones are pretty much 
the only thing you can use.

If you want to distinguish by a burst of audio somewhere in the middle 
of a call, how much audio do you have to work with? If its a fairly 
short burst, looking for a stationary average energy level over, say, a 
couple of hundred milliseconds will tell you it isn't voice. It won't 
confirm that it is FAX, though. It could be something like busy tone

Do you need to distinguish between FAX and other modem signals. If you 
don't care about other modem signals and you have a long tract of audio, 
you can try decoding the V.21 HDLC bursts. If you can decode one between 
pages its fax.

Hunting for the T.30 spec won't get you very far. It specifies the 
message protocol, and not the modem signals. The modem signals are V.21, 
V.27ter, V.29, V.17 and maybe V.34 for the newer FAX machines. All these 
specs (including T.30) can be obtained from www.itu.int. You can 
download a few (3, I think) for free.

Regards,
Steve

Hi all,
First, thanks to Alex, Steve, John and Mr.Vladimir for their 
replies(late reaction I guess? but better late then never). 
I was testing the rolling off of the spectrum of speech myself (on MATLAB)
and it seems to solve my purpose fine.  

Now I have moved on to the problem of finding out which coder has
been used for the speech data.

NOTE: The problem was that I am getting some data along a channel. I have
no contact with the transmitter. First I had to determine whether the data
is speech - which I have done. Now I have to distinguish between the coders
possibly used eg LPC-10 or CELP.

Can anyone tell me possible ways to approach this problem?

All ideas are appreciated.
bye

Now I'm confused...

How could you already have determined whether the data is speech by
looking at the spectrum when it has been encoded by LPC-10 or CELP?

Speech which has been encoded will have a totally different spectrum,
resembling that from a fax or modem transmission.

John

>Now I'm confused...
>
>How could you already have determined whether the data is speech by
>looking at the spectrum when it has been encoded by LPC-10 or CELP?
>
>Speech which has been encoded will have a totally different spectrum,
>resembling that from a fax or modem transmission.
>
>John

Hi 
A valid point John. But going by what I have read about LPC (which i have
to say is very limited) , doesn't LPC cause spectral matching ie the peaks
or the formants in the speech are enveloped by the LPC spectrum. Now i did
my tests of the spectrum rolling off on direct speech signal, not coded
but i figure if LPC does spectral matching then the LPC spectrum should
also roll off at the higher frequencies. 

I ain't no expert about this so I would appreciate if you (or anyone 
else kind enough) would give me clarification in two points:

1. Whether I am right about LPC matching the spectra of speech
2. Would LPC or CELP mask this basic property of speech spectra 
   rolling off? In other words, why do say that encoded speech will
   not show this characteristic ?

Thx in advance.

PS: I hope u are not frustated with all these questions i keep asking.
I am v.v.new to DSP so, i am counting on u guys to help me.