Question about audio artifacts; WMA vs. MP3

Hi:

Why is it that MP3s needs some amount of encoded audio in order to
have any audio at all, while WMA can simply make its own audio? [To
better understand this question, read below]

I have Adobe Audition 1.5 in which I do audio experiments.

Below is my first experiment:

1. I make a silent 44.1 KHz-sample-rate, 16-bit-resolution, monaural
wave file that is 4 seconds long. [To do this, go to "generate" and in
the drop-down menu click "silence". A small windows pops up giving the
number of seconds, I put it at '4']

2. I save it as "silent.wav."

3. I then convert this wave file to a 44.1-KHz-sample-rate, monaural,
20kbps WMA file -- "silent.wma."

4. I close silent.wma

5. I open silent.wma and convert it to a 44.1 KHz-sample-rate, 16-bit-
resolution, monaural wave file and save it as "silent.wav" again --
this overwrites the original "silent.wav."

6. I then close silent.wav and then re-open it. Then I convert it back
to 44.1-KHz-sample-rate, monaural, 20kbps "silent.wma" file
[overwriting the original "silent.wma"].

After generating the silent.wav file I repeat steps 2-6 at least 4
times. Now when I play silent.wma I notice audio in the file that
resembles the characteristic artifacts of WMA.

In my second experiment, I do the exact same thing, except I use MP3
instead of WMA:

1. I make a silent 44.1 KHz-sample-rate, 16-bit-resolution, monaural
wave file that is 4 seconds long. [To do this, go to "generate" and in
the drop-down menu click "silence". A small windows pops up giving the
number of seconds, I put it at '4']

2. I save it as "silent2.wav."

3. I then convert this wave file to a 44.1-KHz-sample-rate, monaural,
32kbps MP3 file -- "silent2.mp3."

4. I close silent2.mp3

5. I open silent2.mp3 and convert it to a 44.1 KHz-sample-rate, 16-bit-
resolution, monaural wave file and save it as "silent2.wav" again --
this overwrites the original "silent2.wav."

6. I then close silent2.wav and then re-open it. Then I convert it
back to 44.1-KHz-sample-rate, monaural, 32kbps "silent2.mp3" file
[overwriting the original "silent2.mp3"].

After generating the silent2.wav file I repeat steps 2-6 more than 4
times. No matter how many times I repeat 2-6, silent2.mp3 still
remains completely silent. Why is this?

It seems like WMA can recognize even the smallest amount of EMI/RFI
and encode it. MP3, OTOH, needs sounds to be louder in order for it to
recognize it.

Just because the file contains 'silence' does not prevent extremely
extremely weak wattages of electrical disturbances from showing up in
the audio file. At some level there is always some amount of EMI/RFI.

>From what I guess, WMA has the ability to encode such small voltages
[resulting from minute electrical disturbances] while MP3 doesn't.
This is probably because of the way the WMA compression scheme is
designed vs. the compression scheme of MP3.

I could be so wrong though.

Please note that there is a huge difference between *bit*-rate and
*sample*-rate. They are two totally different things. Please don't
confuse them.

It really frustrates me when *bit*-rate and *sample*-rate are thought
of as the same thing.

In no part of my experiment do I ever change the sample rate.

I've done the same experiment with audio applications other than Adobe
Audition -- e.g. Wavelab. The results were the same. Adobe has little
-- if anything -- to do with it. AFAIK, it's got most -- if not
everything -- to do with MP3/WMA compressions themselves.


Thanks,

Radium

P.S. In my post I am describing lossy WMA compression. The standard
WMA.

Radium wrote:

> Why is it that MP3s needs some amount of encoded audio in order to
> have any audio at all, while WMA can simply make its own audio? [To
> better understand this question, read below]
> 
> I have Adobe Audition 1.5 in which I do audio experiments.
> 
> Below is my first experiment:
> 
> 1. I make a silent 44.1 KHz-sample-rate, 16-bit-resolution, monaural
>    wave file that is 4 seconds long.[..]
> 3. I then convert this wave file to a 44.1-KHz-sample-rate, monaural,
>    20kbps WMA file -- "silent.wma."
> 5. I open silent.wma and convert it to a 44.1 KHz-sample-rate, 16-bit-
>    resolution, monaural wave file and save it as "silent.wav" again [..]
> 6. [..] Then I convert it back to 44.1-KHz-sample-rate, monaural, 
>    20kbps "silent.wma" file 
> 
> After generating the silent.wav file I repeat steps 2-6 at least 4
> times. Now when I play silent.wma I notice audio in the file that
> resembles the characteristic artifacts of WMA.
> 
> In my second experiment, I do the exact same thing, except I use MP3
> instead of WMA: [..]
> 
> After generating the silent2.wav file I repeat steps 2-6 more than 4
> times. No matter how many times I repeat 2-6, silent2.mp3 still
> remains completely silent. Why is this?
> 
> It seems like WMA can recognize even the smallest amount of EMI/RFI
> and encode it.

Which EMI/RFI? There was no analog step in your procedure.
WMA is probably more prone to rounding errors creeping into
the audio data than MP3 (at least when using the encoders 
and decoders that you used).

> MP3, OTOH, needs sounds to be louder in order for it to 
> recognize it.

MP3 is just working as expected, that is.

> Just because the file contains 'silence' does not prevent extremely
> extremely weak wattages of electrical disturbances from showing up in
> the audio file. At some level there is always some amount of EMI/RFI.

No no no. As long as you just calculate numbers no EMI or RFI will
influence the results (on a properly working computer).

bye
Andreas
-- 
Andreas H�nnebeck | email: acmh@gmx.de
----- privat ---- | www  : http://www.huennebeck-online.de
Fax/Anrufbeantworter: 0721/151-284301
GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc
PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc

On Aug 16, 11:45 pm, Andreas Huennebeck <a...@gmx.de> wrote:

> Radium wrote:

> > It seems like WMA can recognize even the smallest amount of EMI/RFI
> > and encode it.

> Which EMI/RFI? There was no analog step in your procedure.
> WMA is probably more prone to rounding errors creeping into
> the audio data than MP3 (at least when using the encoders
> and decoders that you used).

> > MP3, OTOH, needs sounds to be louder in order for it to
> > recognize it.

> MP3 is just working as expected, that is.

> > Just because the file contains 'silence' does not prevent extremely
> > extremely weak wattages of electrical disturbances from showing up in
> > the audio file. At some level there is always some amount of EMI/RFI.

> No no no. As long as you just calculate numbers no EMI or RFI will
> influence the results (on a properly working computer).

Okay. Sorry. Electrical interference [an analog entity] is now ruled
out.

So just what gives WMA the ability to produce its own audio -- in the
manner which I described in the 1st post of this thread? Why can't MP3
do the same? What is the difference between MP3 and WMA that causes
this characteristic? Or is M$ going to keep it top-secret?

Radium wrote:
...
> So just what gives WMA the ability to produce its own audio -- in the
> manner which I described in the 1st post of this thread? Why can't MP3
> do the same? What is the difference between MP3 and WMA that causes
> this characteristic? Or is M$ going to keep it top-secret?
> 

It may well be Audition's fault.  In its earlier life as Cool Edit Pro, 
the designers "invented" a scurrilous pseudo-floating-point file format 
(which they promptly set as the "default" format) which in terms of the 
header was identical to an integer format - they deduced it was floats 
by inspecting the samples. By default, when the file is silent, they 
decide it is floats - I did exactly this test years ago to demonstrate 
the nonsense that it is. All that remains is for some floating-point 
numerical errors to creep in (wouldn't put it past them), and voila.  Of 
course, it could also be a bug in the WMA encoding - the developers 
probably did not expect whole files of silence!

The only legit floating-point WAVE file is the "Type-3" variety, 
so-called because the dwFormat flag in the header has the value 3 rather 
than 1. Believe it or not, the dodgy 32bit float format's range was not 
+-1.0, but +- 32768.0.

Richard Dobson

On Aug 17, 11:53 am, Richard Dobson <richarddob...@blueyonder.co.uk>
wrote:

> Radium wrote:

> > So just what gives WMA the ability to produce its own audio -- in the
> > manner which I described in the 1st post of this thread? Why can't MP3
> > do the same? What is the difference between MP3 and WMA that causes
> > this characteristic? Or is M$ going to keep it top-secret?

> It may well be Audition's fault.  In its earlier life as Cool Edit Pro,
> the designers "invented" a scurrilous pseudo-floating-point file format
> (which they promptly set as the "default" format) which in terms of the
> header was identical to an integer format - they deduced it was floats
> by inspecting the samples. By default, when the file is silent, they
> decide it is floats - I did exactly this test years ago to demonstrate
> the nonsense that it is. All that remains is for some floating-point
> numerical errors to creep in (wouldn't put it past them), and voila.

As I said before, I tried my WMA/MP3 experiment with many different
audio softwares. The results were the same.

> Of
> course, it could also be a bug in the WMA encoding - the developers
> probably did not expect whole files of silence!

This is probably more correct. It most likely has to do with the
mechanisms of WMA/MP3 compressions than the audio softwares used for
their conversions.

Radium wrote:
..
> 
> As I said before, I tried my WMA/MP3 experiment with many different
> audio softwares. The results were the same.
> 

Could be piled-up dither?

Richard Dobson

On Aug 17, 3:45 pm, Richard Dobson <richarddob...@blueyonder.co.uk>
wrote:
> > As I said before, I tried my WMA/MP3 experiment with many different
> > audio softwares. The results were the same.
>
> Could be piled-up dither?

Don't MPEG Layer 3 encoders usually include an adaptive
quantization step?  If so, any noise or small signals
below the first quantization level will effectively be
rounded to silence.  If this level is above the error
produced by an encode/decode cycle, then silence will
remain silence.  If not, then any compression noise may
well be propagated.


IMHO. YMMV.
--
rhn A.T nicholson d.0.t C-o-M