Part of this study refers to the redbook upper frequency cut.
Also, as I understand, part of this study refers to the ABX
methodology.

In my opinion, ABX tests and the statistical methodology that
accompanies them are NOT DIRECTLY questioned in this experiment.

Nevertheless, ABX tests are used to try to establish a difference or to
establish the absence of a difference (similitude) among the way two
reproduction machines can reproduce music. So we use sound samples as
means of conducting these tests, then we proceed to the conclusion (are
they similar or not).

We knew _before_ this experiment that :
- Frequency samples cannot be used to distinguish two reproduction
machines using ABX tests.
Explanation: A music message is a sum of primary (for instance:
sinusoidal) sounds. This is similar to the way a D/A conversion works.
One could think that if we can distinguish a difference using two
frequency samples, then we can conclude that we can -logically-
distinguish the resulting (as a sum) music.
In fact, several tests (I participated myself in one of them) proved
that we CANNOT distinguish two frequency-based sound samples reproduced
by two different machines, even if we could distinguish (using the same
ABX tests) these machines using normal (time-based) music samples.
Why ? Simply because a human ear is NOT a linear measurement
instrument. Not really a big discovery, some may say. Exactly:
Oscilloscopes and other measurement instruments are much better than
our ears.
Conclusion: Never use frequency-based (primary) sounds to proceed to
ABX tests: The results are irrelevant, the tests are void.

But we have always thought that:
- Duration samples (complex, but of a small duration) could be used to
distinguish two reproduction machines using ABX tests.
This experiment concludes that duration samples cannot be used to
proceed to ABX tests, because the results _may_be_ irrelevant.
Well what then ? What does it mean -may be- ? They are, or they are not
?
In fact, as ABX tests use mathematical (statistical) methodology, the
range rule apply: the measurement means MUST be absolute. No -maybe- is
accepted, or the methodology is broken. Statistics need facts to work,
not approximations.
So, in fact, as this experiment establishes an approximation (or even
-the not measurable possibility- of an approximation, if you are
optimistic), the samples cannot be used anymore. Simple mathematics.

So what ? No problem: It's enough to use longer, complex samples to
validate the ABX tests, that's all. No problem, as I stated.
On the other hand: any ABX test that doesn't use long, complex samples
is void. We can't manipulate maths as we want, pity. :D


-- 
Themis

SB3 - North Star dac 192 - Denon 3808 - Sonus Faber Grand Piano Domus
------------------------------------------------------------------------
Themis's Profile: http://forums.slimdevices.com/member.php?userid=14700
View this thread: http://forums.slimdevices.com/showthread.php?t=54077

_______________________________________________
audiophiles mailing list
[email protected]
http://lists.slimdevices.com/lists/listinfo/audiophiles

Reply via email to