Re: What's the use of a 192 kHz sample rate?

On May 3, 10:34=A0am, rickman <gnu...@[EMAIL PROTECTED]
> wrote:
>  =A0I am simply saying that you shouldn't judge what others can
> perceive by what you can or even what the general public can according
> to "proper tests".

what *should* we base it on?  what the Monster Cable people tell us
that they can hear?

by "proper tests", i am convinced what they mean are truly blind tests
where a sufficient number of subjects, with "good" hearing, are called
on to say if they can hear a difference between two possibly different
sounds or not.  rather than "ABX testing", i think it should be "AB
testing" where the subject hears groups of two sounds, and for each
group the only question the subject must answer is if he/she thinks
the sounds are different or the same (*not* which one sounds "better"
or "less distorted or noisy", etc.).  the test would include an equal
number of identical pairs so that we can unbias it from the subject's
pre-conceived biases.  for each subject you would subtract the number
of false positives from the number of true positives and also subtract
the number of false negatives from the number of true negatives.

now, long ago, when DSD was just coming out, i participated in
something like this that an audio guy (now an author) named Bob Katz
did.  this was before Pro Tools HD and they were using some expensive
system from a company called Sonic Solutions that could do 192 kHz.
they recorded at 192 kHz some test sounds (including some high
frequency percussive sounds like castenets, cabasas, cymbals) along
with synthesized bandlimited (to 96 kHz) tones of all sorts of
waveshapes at a variety of frequencies from below 10 kHz up to 90
kHz.  i don't remember all of the test sounds, but they made sure that
most, if not all, had content well above 30 kHz.  the subject listened
to those sounds in two different forms, but both with a 192 kHz
playback rate.  one form was the raw recorded sound, the other was
processed through a phase-linear FIR filter with a lot of taps (i
thought it was around 300 something taps, it was not a real-time FIR
filter but proceessed one sound file into another) that was flat to
within 0.01 dB up to 20 kHz, had a smooth transition band from 20 to
22 kHz, and then was down by more than 130 dB for 22 kHz to 96 kHz.
we padded the beginning an end of the original sound file with zeros,
of half the length of the FIR on both ends and the non-real-time FIR
was not "causal" and had a delay of 0 (so it started responding before
the first non-zero samples) and the filtered sound file was lined up
in time with the original.

now this isn't what Bob Katz did, but it is what i wished he did:
called the original sound "A", and the LPFed sound "B".  so sometimes
the subject hears AA, sometimes AB, sometimes BA, and sometimes BB
(all four permutations exist in equal quantity) and for each pair, the
subject simply has to say if they think the sounds are the same or
different.  every time a subject says that AA or BB sound different,
we count that as a false positive and subtract that count from the
number of times the subject says that AB or BA sound different.
likewise for false negatives.

now Bob didn't do that, but he did something like it (i think it was
ABX) and there was no statistically measured difference.  people could
simply not reliably tell if the stuff above 20 kHz was removed or
not.  they could not tell at all.  no one could.

now, if they cannot tell if the content existed above 22 kHz or not,
is there a need to have it there in storage or in transmission?   if
there is no need to have it there, and it is removed, what does the
sampling theorem tell us regarding sufficient sampling rate?

> I'm not trying to "prove" anything. =A0I am presenting information which
> you can consider and believe or can ignore. =A0But you can't say my
> statements are false unless you have some information to "prove" they
> are. =A0Human hearing is not a microphone connected to an amplifier.
=A0It=

> is a very complex process which even includes the brain and we
> certainly don't understand it completely.

but we can't hear anything above 20 kHz.  even with percussive sounds
with sharp attacks.  and if we cannot hear anything above 20 kHz, then
40.0001 kHz sampling rate can store all of the information we need.
for practical reconstruction purposes, 44.1 and 48 kHz are sufficient.

now, in *processing* sounds with some nasty non-linearities in the
process, it very well may be necessary to upsample to 192 kHz or
higher to do that non-linear processing, and when it is done, LPF to
20 kHz and downsample back to 48 kHz.  but, except for experimental
purposes, 192 kHz storage or transimssion is not necessary.

r b-j