From: "Graham M Smith" <>
>
> Bruce,
>
>
>>> You might want to read through this:
>>> http://www.minidisc.org/atrac_breakdown.html
>>> to see if there are any signals listed that are anything close to
>>> the sounds of a bat.
>
>
> Thanks, I don't really have the skills to work this out. As has been
> discussed, I'm not really recording the bat sounds, but sounds that have
> been modified by the bat detector down to much lower frequencies than the
> original calls, at least a 10th of the original frequencies.
>
> However, that isn't to say that there aren't some specific distortions
> somewhere in the range as a result of the compression. But I will have to
> work on that on:-)
I went off and read that page. It's kind of the blind leading the blind.
I saw a few of the instances that have a slight chance of being caused
by the ATRAC compression, but most of what's being discussed is
somewhere else in the chain of hardware involved. This is a common
pattern, something goes wrong with a recording or it's playback, and the
inexperienced immediately blame the compression. There is a large amount
of this folklore, and it's the basis of nearly all negativity about ATRAC.
Most of the page is comments that resulted from the graphic near the top
of the page. This graphic is presented as a ATRAC error, it's not. Those
of you new to recording go look at it. If you ever see such a thing in
your own recording's waveform, form your hand into a cocked gun, with
your other hand turn the "gun" to point at your heart. Now fire the
"gun". You have just "shot" what's to blame for that bad waveform in the
heart. This waveform is digital clipping, it may manifest itself in a
number of ways, but this is a common way. Clipping is caused by feeding
some part of your equipment too strong a signal. Typically you have set
the gain too high on the recorder, but it can be in other parts of your
signal chain. It is preventable, and bloody hard to fix afterwards.
Don't trust the metering on your recorder without testing it. Get out
and deliberately record at higher and higher gain and meter indication
and develop a good understanding of what meter indications mean clipping
is starting. Metering is deliberately slowed way down so that we can
read it. That means peaks may not be accurately metered at all,
particularly sharp ones. If your metering has a peak hold function that
may help as it hold the peaks long enough for you to see them. Even
there, most peak hold setups are indicating the average of a few
samples. In the end you have to learn to use a combination of intuition
with metering to deal with this problem. The setting of gain on a
digital recorder should be done differently than it was for analog,
where clipping was not as big a disaster. Always set your gain to have a
pad to handle all peaks. Digital clipping is a zero tolerance thing, you
don't want a single instance of it in your recording.
Since compression is a religious issue, as I noted in a earlier post, I
downloaded the original .wav file provided on the page above. Well, I'm
certain it's not the original that was fed into the MD. It's at a 24khz
sample rate and 8 bit depth. From that I assume that the MD was fed a
48khz sample rate original, and one can hope at 16 bit. That means at
least two resamplings to mess things up. However, I pressed on, I set
that original to loop over and over and recorded it via USB digital to
my HHb Portadisc (Sony ATRAC 4.5). Then using the same USB digital
connection I ran the resulting recording back into my mac for analysis.
I had recorded 12 repeats of the original sample. I then proceeded to
manually scan the entire length of the waveform produced by all 12
repeats. Nothing at all like the bad waveform on the page, only stuff
like the illustrated original. Here's a representative picked at random
from the thousands available:
http://frogrecordist.home.mindspring.com/naturerecordists/mdtest1.jpg
The clipping in the page's graphic, characterized as a ATRAC error, is
not repeatable when you record properly.
Graham, you noted that there were objections in the bat research
community to using MD. I've tracked the origination of a lot of this,
and it generally leads back to one publication. This publication is
still up on Cornell's website as one of their main equipment
recommendation pages. A while back I finally got interested enough to
repeat the test that was supposed to be showing the effect of ATRAC. It
was another case of driving the recorder into clipping. I had no trouble
reproducing the graphic from that page by driving the Portadisc 10-15dB
above the clip level. You can see the offending publication here:
http://www.birds.cornell.edu/lns/recordingnature/html/recordingnature_teche=
quip2.html
Here's my reproduction of the graph at the bottom of that article:
http://frogrecordist.home.mindspring.com/naturerecordists/Clip_6.5_7.5.jpg
The first half is the original signal, the 2nd half is the portadisc
driven into clipping.
Here's what happens if you reduce the gain enough not to clip:
http://frogrecordist.home.mindspring.com/naturerecordists/MD_6.5_7.5.jpg
This sample is in the same format as before, the first half is the
original, and the second half is how ATRAC messed it up.
The internet is a good source of misinformation as well as information.
Telling the difference is not always easy.
Graham, you asked about books to study. If you are to evaluate a digital
system, you first need to know at least something about what it's doing.
A book I'll recommend is "Principles of Digital Audio" by Ken C.
Pohlmann. This will not only fill you in on the details of uncompressed
audio, but has sections on each of the major compression types with
detailed info. That includes ATRAC. The same book also details how to go
about accurately evaluating such encoders. Many people could benefit
from reading all this material.
I do also highly recommend Bernie's book. But it tries to stay
non-technical.
The short description of what ATRAC does is that it selectively
increases the quantization noise in the recording in order to reduce the
bit usage. Increasing it the most where we are least likely to hear it,
and the least where we are most likely to hear it. On a waveform
graphic, we are talking tiny irregularities introduced into the waveform
caused by decreased bit depth in parts of the waveform. Not big hairy
clipping.
Walt
________________________________________________________________________
________________________________________________________________________
|
