From: Dave <>
>
> On Sun, Jul 18, 2004 at 12:41:31PM -0400, Walter Knapp wrote:
>
>>Now, remember that DC offset? Well, it's also a nice way to introduce
>>extra noise in a recording. Digital equipment is designed for signals
>>that are symmetric around the zero line, and will produce noise from the=
>>DC offset. It's why good sound editors have the ability to recalculate
>>the samples to remove the offset.
>
>
> I'm not sure I understand why this would be the case.
There is a good reason why DC offset filtering is in so many pro level
sound editors. And DC offset removal chips exist. It has a number of bad
results.
A few I've seen reported, hardly a inclusive list, but this is not stuff
I invented:
Dynamic range reduction, this should be easily obvious. But I bet it
isn't. It can be obvious after applying DC offset removal. The addition
is at the quiet end.
Noise in the lower frequencies, generally inaudible, but may be audible
if the DC offset is large enough. Often you only are aware of it after
you hear the results of using DC offset removal.
Many filters, including compression which is, after all, a sound
processing filter, produce errors due to it. Or, more correctly, changed
sound.
At the listening end it produces a steady current through your speakers.
They are not designed for that and can burn out the voice coils if the
DC offset is enough. I find this one hard to believe if it came only
from a offset in the signal, I'd think it would take some other hardware
problem. However, the signal DC offset is making the speaker cone
vibration somewhat less symmetric than it or the speaker enclosure was
designed for. The rest position of the cone will not be the designed
one. A sound reproduction error.
Pops and clicks when material is recorded to CD, or even just encoded in
a different sound format. There is a DC transition at the beginning and
end of each file if it contains a DC offset. This may be one of the
click causes in the problems people are having recording CD's.
Particularly likely if they used analog means to get the recording into
the computer where the DC offset can be greater.
I was very skeptical about any value from a DC offset filter when I
first ran into them over a dozen years ago. But, I know how often
applying logic only to a problem can be wrong. So, I tested the DC
filter on a variety of digital audio, not only minidisc recordings, but
even some commercial CD's, some DAT stuff, some digitized cassette
stuff. It was amazing how often you could hear the improvement in the
sound from all those. The sound would simply be clearer.
Nowadays I use a pure digital path into my computer, but I still use the
DC offset filter before other processing. And, though very slight, I
still hear changes for the better in the material at least some of the
time. In the case of Raimund's sample I can see it on the small scale
waveform, which is much worse than I've seen in digital transfers, and
looks like that of a analog transfer. The DC offset in digital transfers
is usually so tiny as to not be visible on the display.
I don't really care if folks believe or not, I only report one of my
tricks for better sound. Try it for yourself.
>>And, if you get closer, lower the gain appropriately to avoid clipping.
>>It is actually the point of getting closer, to lower the gain so less of=
>>the background noise is picked up. It's not a argument against ATRAC
>>that you can force a minidisc to clip, you can force any digital
>>recorder to clip. It has a long history in anti-minidisc, it's Cornell's=
>>main argument to this day. And is as wrong from them as here.
>
>
> I can't find the URL this morning, but I recently read a document of thei=
rs
> on field recording equipment, and the argument in that article was that
> ATRAC (and presumably mp3 and other lossy encoding) was designed around
> *human* perceptual shortcuts. For example, they claim, if two tones clos=
e
> in frequency are present but one is significantly louder than the other,=
> we won't percieve the softer tone and ATRAC exploits this phenomenon and=
> drops the tone of lower amplitude. But if other species can resolve two=
> tones where we cannot, they may be likely to produce calls with that
> feature and any resulting ATRAC recording will not reveal that fact.
> However, the article contained no citations or references to support
> that claim.
This all evolved out of an article on recording tropical birds that
Cornell has had on their site for a very long time, since near the
beginning of minidisc. They still have it on their site to this day, and
I assume that's what you saw. I first saw it before I bought my first
minidisc. I came to the conclusion at that time that they must be
reporting a problem that occurred only with the first versions of ATRAC,
as that's what they tested. The ATRAC 4 at that time clearly did not
make the artifacts they were showing. They were easily bad enough to
completely dominate a recording. Later I found out it was much worse
misleading. Even the old ATRAC they tested would probably be fine on
that test.
All their proof of this "problem" hinged on a test they did where they
used a sound sample that contained two continuous tones at 6.5 & 7.5
kHz. They showed a sonogram that showed the ATRAC producing tone bands
spattered all up and down the frequency spectrum. And stated that
clearly ATRAC could not reproduce close frequency sounds. I'd agree if
it actually did that, read on.
A few years back I happened to be reminded of that test, by yet another
novice to minidisc quoting it and horrified that I do scientific
recording with minidisc. Amazing how many folks in their religious zeal
of anti ATRAC want to save me from my sins and assume I'm too stupid to
have investigated it.
I decided to see if I could get those results with the Portadisc. I made
up a sample of the required two continuous tones, recorded a long
section of that to the portadisc, and went over the waveform and
sonogram along the entire recording length. What the Portadisc gave me
was the same exact two tones, as clean as the original. Now I was really
curious, and started trying different settings in the Portadisc to see
if there was any way to reproduce the sonogram they had. It turned out
you could get a recording that reproduced the pattern Cornell had on
their sonogram. All you had to do was bring up the gain until you were
driving the Portadisc's digital section with a signal that was at least
15 dB above the clipping point. There were all the bands spattered
across the frequencies just like Cornell's sonogram. You could vary the
band distribution by varying the gain at that level. If you set the gain
back down where any sane recordist would set it the bands vanished. Note
you also got erroneous patterns in the zone between no clipping and 15
dB of clipping, but they did not look like Cornell's sonogram.
I should point out for completeness sake that I also experimented moving
the two frequencies closer and closer together. All the way to where the
sonogram was incapable of distinguishing the two the Portadisc reliably
reproduced them. So, don't think it was just not close enough together
to make problems. I also tested various level differences for the two,
nope, the Portadisc still refuses to cooperate and stubbornly keeps
doing it right.
I expect you can do much the same thing with any digital recorder,
compressed or uncompressed. It's digital clipping again, in a massive way.
So, what Cornell's high and mighty expert sound engineer had done is
commit a novice error and try and pass that off as a major flaw in
ATRAC. Had a novice done that, he would have been beaten down in a
instant, but this was a sound engineer with a expert reputation. From
what I've seen quoted from him elsewhere, like Raimund he, for some
reason, hated minidisc, and grasped at anything to try and prove it bad,
without critically examining his evidence. I stated my results and
conclusion in this group some time ago, and for a long time the
sonograms I produced were in my website. They are not there now, removed
in a hunt for webspace for important stuff.
When you see someone claiming that ATRAC won't provide the minor
frequency differences, they are doing so out of the influence of
Cornell. That Cornell still has this page up in their equipment
recommendation section as their evidence of ATRAC's unsuitability drops
them to the level of used car salesman. Deliberately misleading
beginners trying to make good equipment choices. And anyone quoting that
source without including what it really says and is about is also
misleading beginners. They are propagating mythology without identifying
it as such to say it kindly.
I frequently use sonograms on my scientific recordings to check out the
fainter calls behind the louder ones, virtually all my recordings are
looked at very critically along their entire length by sonogram by me.
As well as listened to critically by myself and other call experts. And
the louder callers can be very numerous and very loud. I've been doing
that for 8 years now, spending sometimes months at a time in this
analysis work and the fainter calls or call parts are always reproduced
if they were at the site. Finding and identifying those faint calls is a
important part of the scientific work I'm involved in. They are
reproduced in detail.
I'm still waiting to see this myth about ATRAC in action. It's widely
quoted by folks with little or no technical experience in field
recording with minidisc. But those trying to save me from my error using
minidisc long term will find that I consider it amusing, but tragic
folklore. Because I have the experience, I don't set around trying to
think about minidisc, I use minidisc for valuable scientific recordings,
which I carefully analyze for every scrap of information that they
contain. Recordings that are much appreciated by the scientists that I
deal with. Experts on the calls and the animals.
If someone brings up some new "error" of minidisc I first hold it up to
the rather large block of data, my experience. If it's a error that
occurs much more frequently than winning the big lottery, it will have
been in some of the stuff I've analyzed. That it's not there is not the
end, I normally will try to reproduce it, or work out how they got those
results, depending on just how much time I can devote to it. If it's not
reproducible, my scientific training says it's just dropped to trivial
at best. Science works on reproducible results, something only one
person can produce is highly suspect. It's very hard to justify devoting
much of my limited time to unreproducable results.
Like all my equipment, minidisc was carefully researched before buying,
continues to be researched, and what I get is very critically evaluated.
Walt
________________________________________________________________________
________________________________________________________________________
|