Re: [ARSCLIST] A proposed A/B test (was "dumbing down" of Downloaded Recordings)

[Jon Noring <jon@xxxxxxxxxxx>]

Clark wrote:
> Jon Noring wrote:

>> This does not mean the raw transfers should never be "cleaned up" by
>> those interested in doing so. But we should never throw away the raw
>> transfers (so long as they are properly done).

> CJ Should also make them available for people who want the Entire Sound.

Well, I did imply this, but should have made it more explicit.

During the time a few of us were exploring Project Gramophone/Sound
Preserve (and it is still there, in a manner of speaking -- the
copyright issue is a killer), we looked at the mass transfer of large
numbers of 78s. And the archival storage of the raw transfers was a
given -- we were not going to concern ourselves with restoration,
since that is a step separate from transfer, and could be done and
redone anytime in the future. We expect audio restoration algorithms
will continue to improve.

Of course, it is entirely possible that in the future much better
methods for transferring grooved recordings will be developed, which
can outperform the present state-of-the-art of moving coil cartridges,
etc. An example is if ultra-high resolution microtopography (such as
being studied by LBL) can become practical where we can digitally
reproduce the precise geometry of the groove on a 78 recording,
including all imperfections. This allows computer analysis to
interpolate what the groove shape was when mastered (before any
pressing imperfections, groove wear and damage, and the usual
"boulders" sticking out in the groove due to abrasive in the
substrate), and then "play that back."

(The problem with a stylus being dragged through a groove is that it
integrates everything it "sees" at every moment in time. Thus the fine
cross-section information in the groove is reduced to a single "value"
-- a significant loss of information, and irretrievable once lost,
that would otherwise be quite useful. Laser playback, if perfected,
might be able to better utilize the information it sees at every
cross-section of the groove, and in fact this might be the method to
make microtopography practical.)


>> Get several "audiophile" and 78 collectors
>> together to see if they can tell the difference between:
>>
>> A: 78 playback --> pre-amp --> amplifier --> speakers
>>
>> B: 78 playback --> pre-amp --> AD --> DA --> amplifier --> speakers

> CJ If I may say, "the literature" is full of that stuff, wherein no group
> (as a group) has been found to detect any differences. Hence, CD, MP3, etc.
> etc.

Of course!

But it's amazing how many people don't believe this. Thus the
experiment to begin to refute a lot of misconceptions concerning
digital audio in the archival and reissue communities.


>> Provided of course the AD/DA is done at 96k/24-bit sampling using
>> the highest professional grade equipment, and all other equipment
>> remain the same. If so, I assert the blind test will show no
>> discernable difference, and if this is indeed the case, then simply
>> putting a digital storage step between the AD and DA steps will not
>> change the "calculus" since the storage does not alter the bits one
>> iota provided storage is lossless.

> CJ No such thing as "lossless", sorry.

What I refer to lossless storage is that once digitized, one does not
apply a lossy compression algorithm such as MP3. Compression is done
losslessly, so when uncompressed the resultant "wave" is identical
bit-by-bit to the original. Typical lossless compression algorithms
for CD audio yield about a 40 to 50% reduction in file size -- all
audio has entropy limits to how much lossless compression is possible.
Very noisy material will compress even less since random noise doesn't
really compress.

You must have interpreted my use of the word "lossless" in another
context. I was speaking completely in the digital realm, not the
analog realm.


>> (p.s., since I know someone will ask why 44.1k/16-bit is not sufficient.
>> What we really need to do is to accurately represent the noise in the
>> recording, especially impulse type noise, and this does require a
>> higher sampling rate and bit depth to get a fairly good representation
>> of the noise. The analog side of the AD clearly has to be pretty
>> accurate on impulse type noise. I'll let Erik and others common on
>> this since I know little about the real world of state-of-the-art AD
>> converters.

> CJ Recently it has developed that 44/16 can be made to sound far, far better
> than almost anyone has heard, better than anyone had any right to expect.
> This, from an avowed and early foe of CD as-it-was-known-then.

Anyone who understands Fourier Analysis knows that,
theoretically-speaking, 44.1/16 can accurately represent the signal
sufficient for human listening. It's the real-world implementation in
the A/D conversion (at the mastering end) and the D/A conversion (at
the listener's playback end) that can lead to less than theoretical
results.

But note *carefully* what I said about 96/24 being preferred for
archival transfer. It was not because we need that to accurately
reproduce the wanted signal, but rather in order to accurately
reproduce the noise, particularly impulse noise.

This may seem odd since all we are really interested in is the signal
(e.g. the music). But for purposes of audio restoration, more
accurately representing the noise (especially impulse noise) may
confer advantages. Thus, we would be remiss not to take advantage of
96/24. 96/24 can certainly be downsampled to 44.1/16 as needed, so
it's not an either/or situation. I'm not sure anything is gained going
even higher, such as 192/24, though. An area needing studying specific
to 78 transfer and restoration algorithms.

Jon Noring