naturerecordists
[Top] [All Lists]

Re: Samplerate conversion, and ATRAC decompression mystery

Subject: Re: Samplerate conversion, and ATRAC decompression mystery
From: Aaron Ximm <>
Date: Fri, 15 Aug 2003 01:36:55 -0700 (PDT)
Hi Walt,

Thanks for such a considered response, you've caused me to reconsider some
of my assumptions about exactly what I might be hearing.

> > I had two or three ideas about the differences I'm hearing:
> >  (a) the Sony is doing some un-specified post-processing to handle case=
s
> > where it must generate a lot of 'filler' white noise, to insure that it=
's
> > either masked by higher frequency noise, or as non-periodic as possible=
;
>
> It needs to generate no filler. I've not seen any sonograms of any ATRAC
> recording where deliberate filler was introduced. It does not do this,
> it only works to reproduce what it got not invent new stuff.

I had recalled that there was a "masking" interval following transients,
in which the original waveform could be safely discarded and low-volume
noise synthesized to replace it on decode -- as long as a certain ratio of
transient volume to noise volume was maintained... I was supposing that
perhaps what I think I'm hearing might be a periodicity that was making
these (hypothesized) "masked" noise samples more audible.

But this was pretty much uninformed speculation!

> >  (b) EDL is not implementing the ATRAC decode recipe correctly;
>
> Or just differently, note that the difference could also be hardware,

It's the "differently" that comes as a surprise, I'd read somewhere that
the de-code was fixed and inferred that for a given ATRAC block there was
only one legal "decode", no post-filtering, waveshaping, etc. allowed by
the decoder -- whatever "allowed" means.

One interesting thing about the EDL system btw is that there isn't any
clocking, it reads the disk as a data disk and I presume (perhaps
erroneously) that it does this with essentially perfect accuracy -- then
the transfer is lossless over SCSI (it better be!) and the decode is all
done in software on the PC.

> The compression is not even getting into heavy breathing on this stuff.

Of this one point I'm not 100% convinced; I certainly stand corrected on
my inaccurate use of the phrase "white noise" ~ but I intuit that there is
something "complex" about water/wind flap recordings (or even traffic
noise), in which a lot of non-tonal noise seems to approach, to my ear
anyway, genuinely random noise...

Perhaps its time to go back and study some "complex" recordings to hear if
that artifacts I detect are in fact completely dependent on the decoding
device, as they may well be -- it'd be nice to know that some recordings
I've lamented hearing "ATRAC artifacts" (emphasis on quotes) in are in
fact capable of very clean decompression under better circumstances.

> parts, for instance. Differences in exactly what the software really
> does, which is usually different from what we think it does, and even
> different from the official description of what it does. The trickery in

Exactly my suspicion -- as a professional programmer... :)

> Ok, here's a sonogram:
> http://wwknapp.home.mindspring.com/images/Flagsono.jpg

Thanks!!! Remarkable how 'invisible' the differences I are in this, they
sound so pronounced on my monitors -- in particular I hear a higher-
frequency hiss in the Sony decode that I don't hear much at all in the
MDTE version... and I can only just see more energy on the top of the Sony
(if it is in fact the middle).

> slight intensity difference rather than structural. Look at how closely
> the difference part mimics the louder parts of the others.

I had wondered about that, perhaps I will try to hand-normalize them one
to the other (regardless of my a priori assumptions that they should be at
the same level) and listen again.

> synthesizer for the decode. It is known that slight differences do exist
> between the output of the decoders for various versions. Slight

That was my real question to EDL -- I haven't been able to find anything
online about the specifics, I assume it's all under NDA or something.

Which is really too bad, I'd happily pay someone to write a decoder for my
files that "sounds like" the Sony!  In the meantime, I'm back to using it
for sensitive transfers -- sigh. I do so like the 5x speed benefit of the
EDL system... and the fact that it's essentially a drag-and-drop endeavor
to transfer files. :/

  b  e st
   aaron

  
  http://www.quietamerican.org



________________________________________________________________________
________________________________________________________________________

<Prev in Thread] Current Thread [Next in Thread>
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU