18 Jun
2018
18 Jun
'18
7:24 p.m.
Paul Davis <paul(a)linuxaudiosystems.com> writes:
this is more audiophile-level woo.
virtually nobody alive can tell the difference between a good digital
recording and an analog one. those who think they can will rarely (if ever)
agree to double blind testing.
there can certainly be things about analog equipment that are hard or
impossible to capture with a digital equivalent.
but to claim that the recording process is the source of audible
differences? even if this might have been true of the earliest days of
commercial digital recording, it isn't true anymore.
I think one thing hard to keep track of is nonlinear processing. If you
take a triangular signal and lowpass-filter at 20kHz and reproduce
digitally in the proper manner, you will not hear a difference in the
audible frequencies.
If you square that signal or limit it or distort it, the results in the
audible domain will be different depending on whether you do that
operation on the original signal, on the low-pass filtered signal, or on
the digitally sampled low-pass signal.
You can compensate for that if you _synthesize_ your original signal and
take good care of what you do. If what you start with is a _sampled_
signal in the first place (maybe in order to better reflect the
imperfections of the analog implementation) you don't have the high
frequency content/information anymore in order to determine how it folds
back into the audible domain when doing non-linear processing.
I'd expect this kind of effect to be probably audible with 48kHz
samples. Probably a quite less so with 96kHz.
Of course it isn't relevant for mere recording and replay: it's only
non-linear processing that has a direct effect. Well,
actually... Simple low-order _linear_ filter devices usually are
transformed into the digital domain using the bilinear transform, and
the bilinear transform has less linear phase and frequency relations the
closer you come to the sampling frequency. So again, for modelling
simple analog circuits, you are better off using sampling rates that may
seem silly for mere reproduction.
So it's my expectation that reimplementing simple analog synthesizers
(that have waveforms reasonably close to what they are supposed to be in
theory) in a convincing manner will likely warrant investing higher
sample rates than you'd usually consider sensible bargaining for, either
when synthesizing or when working from sampled versions of the original
oscillators.
So there are some reasons that the "pure snake oil" realm is a bit
further with regard to analog synths than one would like to think. Of
course it's still out there somewhere.
--
David Kastrup