From: Gianni Pavan <>
>
> Raimund,
> it is difficult to say something about the spectrogram you show:
>
>
>>http://www.avisoft-saslab.com/compression/MDtest2MD.gif
>>
>>As you can see clearly on the waveform, there is for sure no clipping at
>>that location (the amplitude is much higher at the start of the call).
>
>
> The "missing" harmonic maybe an artifact or maybe it is real.
A pattern of anything like that nature has not occurred in all the years
I've been staring at sonograms of my minidisc recordings. I'd say to be
ATRAC caused would make it extremely rare, with odds of occurring like
winning the big lottery.
I vote for processing artifact of some kind. It covers 0.35 seconds,
that's over 15,000 samples wide. A huge amount of time in sound
processing. ATRAC works in much shorter frames, and has little
connection downstream like that. The premise proposed that the louder
sound just before blanked it out would simply not occur, no connection.
Even as a processing artifact it's interesting, and rare. The sonogram
is very crude, and the explanation may be in there.
> To have a better view of what ATRAC or MP3 do on a recording I suggest to=
> plot the spectrogram with 96dB or more dynamic range: this way it is easy=
> to see all the portions that are cut by the system because "believed"
> unaudible for the human ear. If the recording has a background noise, as=
> most nature recordings have, you'll easily see (and demonstrate) all the=
> "holes" created by psychoacoustic compression.
> I'm sorry I can't now post an example.
Doing this can also induce all kinds of artifacts. Remember the analysis
software is a form of sound compression. It's extremely crude, and has
artifacts much bigger than those of a very highly refined compression
like ATRAC. At one point one of the shareware programs that floated
around would make a sonogram for you, a pretty good sonogram. But, the
fun part was it would also take a sonogram and convert it back to sound.
Fun to play with, but a single round would produce really poor sound
compared to the original. Yes going both ways probably at least doubled
the errors. That would produce obvious holes easily.
I should note that I've managed a few times to produce ATRAC/no ATRAC
soundfile pairs a few times that were in accurate enough sync along
their entire length to produce a quality difference file. In other
words, what ATRAC removed. I did not find a batch of holes, selected
faint calls hidden by louder foreground or anything like that when I
looked by sonogram and listened to the sound in those difference files.
What I did find was that the difference file was a very close duplicate
of either the ATRAC or no ATRAC file, but at a much reduced volume.
Especially obvious if the gain is matched to the original files. In
other words what ATRAC removed was what it left (and what it left was
what it removed, kind of zen like). Exactly what we want a quality
compressor to do if you think about it. Note the only ATRAC I've managed
to test this way was ATRAC 4, not the ATRAC 4.5 I now use.
So, the idea that there are "holes" from ATRAC seems a little strange.
Such holes would be very obvious in a difference file. In my experience
that not only are not obvious, but appear to not exist.
Walt
________________________________________________________________________
________________________________________________________________________
|
