John Campbell wrote:
> I'm curious about the testing procedure. I presume that you made
> compressed and uncompressed copies of an excerpt from a CD. In which
> case, you are dealing with processed signals, though possibly less
> processed in Gordon's case than in some other output audiofiles, as I
> believe he does not mix different source recordings. However, on a
> couple of his CDs that I have heard, there are gain changes and
> tracks are cross-faded. I believe that no equalisation, limiting or
> other DSP is used on these, but is that also true for the sounds you
> used? In late 2000 there were some files on this list for
> examination of the effects of data compression, but these too were
> from processed sources. With these, the extent of the processing is
> anyone's guess.
>
> The point I'm getting to eventually is that unless one compares
> unprocessed original recordings that either have or have not been
> data compressed (and are obviously identical so far as all other
> aspects of the recording chain are concerned), then I'm not sure that
> the test is wholly worthwhile. Simply, the signal at the "finished
> product" stage may have been sufficiently processed or manipulated to
> minimise or neutralise any perceptible differences between
> compresssed and uncompressed audiofiles.
Before you get too stirred up about "unprocessed" vs "processed" it is
well to remember that the mic and preamps process the sound signal and
only give a approximation of what it was. In fact the very environment
out there processes the call before it reaches the mic.
What you are also saying is that a DAT recorded original transferred to
CD is changed by the process. I would agree, everything we do changes
the signal some. Even simple digital transfers are not 100% accurate.
That's why there are error correction routines in the software.
> So, is anyone aware of tests on original recordings having taken
> place? Complicated to set up I know, but probably essential if we
> really want to know what is happening.
They exist, in a sense I do that every time I record and monitor the
input to the recorder. As does any other MD user who monitors the input.
Every comparison has to be studied very carefully to find out just what
it tested, however. You should be aware that in evaluating ATRAC itself,
you have to isolate the effect of ATRAC from everything else. That means
that all the way through to the ATRAC encoder has to be the same as the
path to the "uncompressed". It also means that however you record the
uncompressed, it needs to be absolutely perfect. And finally, the output
end from the ATRAC decoder through to your speakers/headphones or
whatever (including them) must be identical to the path out of the
uncompressed. Ideally this should not only be the exact same design
circuits, but the exact same circuit. Used with the same settings,
headphones worn the same way and so on.
The practicality of doing this has had the meaning that for most testing
what we are testing is one entire system against another, not just
ATRAC. Working out experimental designs that test only the ATRAC leaves
very limited choices. In my case I took advantage of the Sony MXD-D3
deck's ability to play either CD or MD through the same outputs and to
clone CD to MD directly internally. That test, along with a comparison I
did later of the same samples with the HHb Portadisc is here on Doug's site:
http://www.naturesongs.com/walt/difference.html
In that case I used several CD's as source material, primarily from Lang
Elliot and Wil Hershberger. The text explaining things would have to be
found in the posts I put up at the time, the page is only the raw data.
And since the key is given right there it's hard to do it fully blind.
The nature recordists who did the test at the time did no better than
chance at identifying the MD by listening. In fact they did not do all
that well by analysis and I provided sonograms.
Many of the so called comparisons have been between pro level DAT
recorders and walkman MD's. And really say more about the differences
between those two categories of equipment than what recording technology
was in them.
Even once you get all the technical part worked out there is still one
more problem in doing listening comparisons. This has to do with the way
our brain processes sound. Your brain enters in your attitude about the
sound in it's interpretation of what you hear. So, if you think one
sound will be better than another, your brain will make sure you hear it
that way. It's very hard to get around this except by blind listening
tests. It is the main way that we have so many who report hearing
differences in ATRAC encoded material when they know that's what it is.
The same folks, when given a blind test where they don't know which is
which have been shown over and over again to do no better than chance.
I realize there is this belief that something must have changed and if
we listen just the right way, we will hear it and find out just which
samples it threw out and which it kept. I, too know something had to
have changed, the ATRAC threw out 100% of the original data (not some
samples only), generated new data that described the original but took
up 1/5 the storage space, and then used the new data to generate a
reproduction of the original. One that's good enough we can't hear the
difference, and neither can the animals we deal with. I do not expect to
hear the differences, no matter what technique I apply. I do know I can
find small differences by sonogram. I have come to these conclusions as
a result of many years of recording with ATRAC encoding. I did not start
out believing I would hear no difference.
Walt
________________________________________________________________________
________________________________________________________________________
|