richpeet:
> I would like to take you up on the statement that you have complex
> sounds that gives disturbing artifacts in atrac.
We misunderstand each other. I am saying that the process of *manipulation*
reveals the artifacts of lossy compression. As I repitch, time-stretch,
filter, slice, dice, bend, fold, and mutilate, the perceptual encoding
unravels.
Perceptual encoders divide the audio spectrum up into bins, then make
decisions about how roughly to quantize each frequency bin based on a
perceptual model. If you rearrange the relationships between the various
frequencies, you invalidate the perceptual model and the decisions derived
from it, and the quantization distortion in the individual bins may rise
above the masking threshold to become perceptible. The result eventually
starts to bear a resemblance to that "real audio 56k" sound we all know.
Were the material I'm gathering intended for unmanipulated playback, I would
choose minidisc and ATRAC without hesitation. I'm no superstitious
audiophool (Dan Dugan can vouch for my skeptical credentials), and IMO,
ATRAC is very very good. However, my application is unusually demanding on
source material.
> If you e-mail one of these complex sounds then I will convert it to
> a "disturbing atrac" sound and back to pcm wave. I can then e-mail
> it back to you. This will include one analog conversion as consumer
> level machines do not have digital out. It will mean two atrac
> conversions, one in and one out.
>
> Then you tell me which file is yours without looking at a graphic,
> just listen.
Were I to send you a PCM recording of a barking dog, have you ATRAC it and
send me back two files to examine double blind, I have little doubt that I
would fail to discriminate between them reliably, certainly in any normal
listening environment, possibly even on my old mastering rig.
Were I to send you a finished composition consisting of heavily manipulated
samples, I think the odds of my successfully identifying the ATRAC'd file
are increased, though by how much is hard to say. It has often been
observed that perceptual codecs are most effective on acoustic sources, and
that they can fall down when fed artificial source material such as
electronica. The reason that happens is simple: perceptual codecs are
primarily designed and tested using acoustic materials which have certain
universal attributes, while it is difficult for the designers to anticipate
all the possible attributes of all possible artificial materials that may
crop up. As a mastering engineer who was occasionally responsible for
performing extreme manipulations on client materials (e.g. 12 dB of peak
limiting -- ouch) and then trying to squeeze the output into perceptual
codecs, I often had to adjust my settings to optimize everything for the
perceptual encoding, so my own experiences corroborate the popular wisdom.
Robert Orban, designer of the most popular radio processing equipment out
there, went on a public awareness blitz a couple of years ago to let
mastering engineers know just how poorly their material was holding up when
it had to undergo a second stage of extreme manipulation at the broadcast
stage. The best thing to send to a radio station, if you're going for
optimum playback rather than gunning for the attention of the program
director, is a totally unmanipulated source! Now ATRAC is not even in the
same ballpark as what happens in the Optimod at a typical commercial
station, but what what _I'm_ doing is _much_ more extreme than radio
processing, so like Orban, I want to work with unmanipulated source
material.
Were I to send you a finished composition consisting of manipulations
performed on an ATRAC source, the likelihood that I would be able to to
discriminate between the original and the "re-ATRAC'd" version jumps
substantially, because we would be violating the principle of avoiding
"cascaded codecs". But in that case the composition would have been
compromised from the start...
I will illustrate the issues involved with a visual composite. The
following link contains two eagle beaks, one from an uncompressed file, and
one from a JPG saved at 7 out of 10 quality on the Photoshop slider.
Examine the area around the edges of the beaks, and you will see JPG
artifacts on one of them. (The file itself is a JPG saved at 10 out of 10
quality, high enough that the artifacts from that file-save iteration will
not wreck my illustration. I would have saved in an uncompressed format,
but no uncompressed formats offer consistent browser compatibility.)
http://wwww.rectangular.com/eaglebeak.jpg
It is audio analogues of these artifacts that I'd like to avoid.
You may also note that the images are blown up enough for the pixelation to
start to become apparent. Perhaps this sounds strange, but if I couldn't
go PCM I would choose the creative limitations of cassette over the creative
limitations of ATRAC for highly manipulative audio collage work. I am not
partial to quantization distortion, though many other audio composite
artists either like it or don't care (e.g. DJ Shadow). For me, tape hiss
and analog saturation are "f___inpunkrock!" while quantization distortion
and perceptual encoder artifacts are just distracting and annoying.
So now you know why I'm trying to find a good solid unit to put between my
mics and my laptop.
Best,
-- Marvin Humphrey
CD design website - http://marvin.mrtoads.com
________________________________________________________________________
________________________________________________________________________
|
