I apprecciate your detailed comments and respect all possible (
although quite contraversial) resources. But simply said: in MY SETUP
(Mytek AD, Millennia, Schoeps, Neumann etc.) recordings in 96kHz sound
REASONABLY better than in 44 kHz. That´s all. (Only one with no ears
would not hear that). No theory can change it and I have absolutely no
reason to persuade anyone about it. I just wanted to share it for
those who are interested to know. If I have a time, I may post some
samples, but cannot promise it at the moment.
It is a bit more complicated to move in 96 kHz but not everyone feels
like sacrifying beauty of thing to a comfortable life ...
Bye.

Ivo


"Arny Krueger" wrote in message ...
"Ivo" wrote in message
om
"Arny Krueger" wrote in message
...
"Ivo" wrote in message
m

If you didn't do level-matched, time-synched, bias-controlled tests
the evidence of your ears is highly contaminated by evidence from
your eyes, and hearing evidence unrelated to sound quality.

If we are talking about very subtle differences, then maybe your
remark should be considered. But these are not subtle differences at
all, this is clear like night and day ...

You probably don't know how many times people have said this, only to be
embarrassed by their own performance in level-matched, time-synched,
bias-controlled tests. I've been there and done that, more times than I can
count. I've seen dozens of well-meaning people take this particular fall,
including myself.

Why did I get into the PCABX thing? I'd taken this fall way too many times,
but I didn't think I could get away with always saying "no comment" and
still be happy with myself. Curiosity got the best of me. I figured out how
to make a wide range of listening tests with reasonable controls be far
easier, more accessible for as many people as possible, and repeatable.

If you've got a halfways decent DAW, the sample rate issue is already laid
out for you at:

http://www.pcabx.com/technical/low_pass/index.htm


and

http://www.pcabx.com/technical/sample_rates/index.htm


If you tell me that "these two speakers sound different" or "these two mics
sound different" without out a carefully-run listening test, I'm prone to
believe it. I've been there done that many times with and without the
controls, and that sort of comparison rarely comes out sounding the same,
even with every conceivable reasonable control in place.

If you tell me that "these two good amps sound different" or "these two
sample rates 38 KHz sound different" without a carefully-run listening
test, I'm prone to be skeptical. I've been there done these many times with
and without the controls. This sort of comparison rarely comes out sounding
different with the even just the basic controls in place. I've seen it done
by enough different people and with enough different systems and samples
that it isn't just about me.

If you search the RAP archives you'll find people like Massenberg and Katz
saying the same basic thing; results, controls - your whole basic DBT and
sample rate enchilada. Now, try to tell me that they don't have the proper
resources at their disposal to evaluate this issue!

Norbert Hahn wrote:

Codifus wrote in message .net...


My suggestion is simply based on the math. To make an 88.2 file become a
44.1, you simply divide the number of samples by 2. For a 96 file to
become a 44.1, you have to divide the number of samples by 2.176871.


What does division by 2 or 2.17 mean when it comes to sample rate
conversion? If you devide the numbers contained in a wav file by 2 you
simply reduce the level by approx. 6 dB. The sample rate doesn't
change.

Sample rate conversion is performed in serveral steps and several
algorithms are used he

* To convert 88.2 kHz to 44.1 kHz the samples must be lowpass filtered
such that no frequency higher than 22.05 kHz are contained. After that
you simply take every other sample and copy that to the new file.

* To convert 96 kHz to 44.1 kHz you must first upsample to a frequency
that is an integer multiple of both 96 and 44.1. Which is 147/640.
This is simply done by adding 146 zeros after each input sample. Next,
low pass filtering to 44.1 kHz must be done. Last step is to pick
every 640th sample and put it into the output file.

In both steps the quality of the low pass filter is essential. A
deeper discussion can be found in http://www.analog.com searching for
"Engineer To Engineer Note" EE-183.

Norbert
Thanks for the refernce. I read the article, the math was a bit too
intense for me, but I got the jist of it. Now, is it safe to assume that
all SRCs convert the same way? My SRC is software, CoolEdit 2K to be
exact. I'm not sure how CE2K converts,none of the manuals or help files
describes how it works, but if it did convert down from 96 to 44.1 as
the analog devices SRC does, I would record at 96K as well. I'm pretty
sure it does not, though, because the time taken to convert from 96 to
44.1 is similar to 88.2 to 44.1. And besides, since the analog hardware
is built to do the math, it's bound to be much much faster than to SRC
in software, therefore it can upsample and then down sample without too
much of a time penalty.

CD

Codifus wrote:

Thanks for the refernce. I read the article, the math was a bit too
intense for me, but I got the jist of it. Now, is it safe to assume that
all SRCs convert the same way?

I don't know about older software and I'm far from knowing all SRC software.

My SRC is software, CoolEdit 2K to be
exact. I'm not sure how CE2K converts,none of the manuals or help files
describes how it works, but if it did convert down from 96 to 44.1 as
the analog devices SRC does, I would record at 96K as well. I'm pretty
sure it does not, though, because the time taken to convert from 96 to
44.1 is similar to 88.2 to 44.1.

The SRC software in CE is old but performs more or less the same steps
as described in the paper. I use CE Pro which is now Adobe Audition and
found that SRC performs quite well sonically but takes a lot of time
because CE first does the required low pass filtering over the whole file
then rereads the file for decimation, does some post filtering, and finally
the peak file (which is used for the screen image) is rebuilt.

The reason why I use 96 kHz in CEP is that some algorithms perform better
at higher sampling rates, notably generation of reverb. I hear a big
difference when reverb is calculated at 96 kHz rather than 44.1 kHz.
Going to samples rates higher that 96 kHz make no audible difference for
me so I stick to that.

Please note that better reverb is not due to the higher sample rate as such
but due to the specific implementation of that function in CE. There are
other functions that perform "better" at lower sample rates - click removal
for example.

And besides, since the analog hardware
is built to do the math, it's bound to be much much faster than to SRC
in software, therefore it can upsample and then down sample without too
much of a time penalty.

Yes. SRC done in a DSP is pretty fast while CoolEdit uses a modular approach
by performing one step after the other (no overlap). CE allows to control the
low pass filtering. The SRC algorithm is used in other functions of CE as well:
changing pitch or changing speed.

Norbert