[Scott Dorsey]

karle wrote:
Ok I will re-explain in detail then.

What I want to do is:

1- Record a performance with 2 microphones fed into a console
2-Take the console outputs and send these to 7 identical recorders (computer
/ AD converter, or maybe standalone recorders), each recording at a
different sampling rate
3- Play back the various recordings to test subjects to see which recordings
they prefer
4- Then run the experiment again with X brands of converters, but each brand
recording at only at the most preferred and the least preferred sample rate.
5- If, among all the brands, the most preferred sample rate remains, then I
can assume that the distortions and artifacts introduced by the converters
aren't as meaningful as the number of samples recorded per second. Because
the filter slopes and converter performance will vary from manufacturer to
manufacturer.
I will then concentrate my efforts on the ultrasonic perception part.
6-If, among all the brands, the most preferred sample rate differs, then I
need to dig deeper.

What are the flaws in this plan?

You still don't know what you're measuring when you do it. You have
preferences but you have no idea _why_ anyone has a preference.

See... in a perfect world, the ONLY difference between different sample
rates is due to bandlimiting. If you use a higher sample rate, the
frequency response is extended.

But, we don't live in a perfect world, so sometimes there are artifacts
that are not due to bandlimiting. For example, with the Panasonic SV-3700,
there is a dramatic difference in sound quality between 44.1 and 48 ksamp/sec
operation. That's not due to the change in sample rates itself, per se,
because it's not due to the bandlimiting. It's because the converters in
the SV-3700 were designed by idiots and have massive aliasing problems,
and those aliasing problems change with the sample rate.

On the other hand, you can take a Prism AD-124, sold at about the same
time as the SV-3700, which was designed by people who knew what they were
doing, and really hear no difference between the 44.1 and 48 ksamp/sec
rates.

So.... do you want to change sample rates in order to tell if people can
hear bandlimiting, or do you want to change sample rates in order to tell
if the converters have other artifacts? Because both of these things
happen at the same time.

My main goal is to establish a transparency "upper bound"
sampling rate, as Bob stated it. If I reach point 6 in my above plan, then I
will try to isolate the various elements.

Folks have done testing for years and years to determing "upper bound"
sampling rates in the real world, and it has changed as the converter
technology has changed. The old DASH recorders ran at 54 ksamp/sec, because
with the converter technology of the day that was the lowest rate they
could run at without clearly audible high frequency problems. Today the
converter technology is much better, and so we can run at much lower rates
without audible differences.

So if you test YOUR equipment with a listening panel, and you record and
play at different rates, the results you get are valid ONLY for YOUR equipment.
They cannot be generallized. If you want a test that can be generallized
you have to be a lot more sophisticated about it. If that's all you care
about, that's fine, but strictly speaking it's not very useful.

Check the test in the recent JAES issue that folks have referred to. It
wasn't last quarter, I think it was the issue or two before that. It was
nicely one.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."