Re: Artifact Audibility Comparisons

"Tom" <tom.derham@xxxxxxxxx> wrote in message
news:10b71c7b-36eb-4e97-8452-ee56fd457673@xxxxxxxxxxxxxxxxxxxxxxxxxxx

A round about way of saying that jitter is modulation
distortion. No carrier, no modulation.

In the case that your original audio signal is a sine
wave, yes it can be considered that way.

The cases where the original audio signal can be correctly interpreted as a
collection of sine waves describes every other continuous signal known to
man, right?

And therefore signal dependent, right?

Dependent on a signal being present, as opposed to being dependent on the
detailed properties of the signal.

Therefore I believe my point about it not being
appropriate to add noise (of any sort) onto an original
audio file in order to simulate an audio file "with
jitter" is valid.

Not always.

I would be interested to know a case where you can
consider it to be additive, given the above.
(note: additive == signal independent)

No additive works when there is no signal present. It's just a trivial case.

Further I believe it is valid to say
that any noise/distortion on the output audio signal due
to having been sampled by a jittery clock is always
signal dependent.

Which is helpful by making a useful distinction between
jitter and other kinds of distortion, how?

In your original waveforms you used additive, wideband,
signal independent, uncorrelated noise, right?

I'm missing the context for this.

So I am saying jitter distortion is not like that.

A spectral analysis is just data.

And without phase it is incomplete data...

Spectral response minus phase response is amplitude response.

and even with
phase it is difficult to intuitively interpret.

Since when are we limited to intuition?

Think of a wide, flat spectrum - if its phase is random
the time domain audio sounds like noise, if its linear it
sounds like an impulse (followed by nearly silence), if
its quadratic it sounds like a longer chirp... and that's
just for trivial cases. The differences are because of
the constructive/destructive summing of #all# harmonic
components over time, which obviously depend on the
initial phase of each component.

I'm lost.

Therefore, just because you see a spur in an audio
spectrum at a certain frequency at -50dB, that doesn't
mean the energy at that frequency is constant in the time
domain, and therefore it doesn't mean that it will
#sound# like a discrete tone (at a low level).

Oh, you're looking for 100% certainty as opposed to 99% probability.

I've been looking at spectral analysis information for about 43 years. I
can't think of one case where I went and did more analysis of a spike, and
did not find something that wasn't in some sense coherent.

Therefore
I am not convinced by arguments that say spurs at minus-
whatever dB must surely be inaudible just because a
single tone by itself wouldn't be at that level. It might
be true,

Your error here is that tones with consistent phase are invariably more
audible than ones with random phase. If you can't hear a simple tone at a
certain frequency, odds are excellent that you won't be able to hear a
modulated tone at the same frequency. For openers, a modulated tone doesn't
have energy at just one frequency. A modulated tone won't create a single
spike, it will almost always generate at least 3 spikes.

but such arguments are insufficient.

Insufficient for what? Insuffiient to make reasonble assumptions?

As an easy example, think of another type of signal
dependent noise - quantisation noise - e.g. from
converting a 24-bit wave file down to 8- bits. If the
original audio is a simple sine wave, then you see many
discrete spurs in the resulting spectrum. But, for a
real music signal with a rich spectrum, this spectrum is
convolved with the set of spurs, creating a total noise
spectrum that looks basically like that of uncorrelated
wideband noise. But it is not, it is correlated with the
signal.

No mystery. If you convolve with something that is
changing, then the result changes with the changes in
the something.

I am talking about a frequency spectrum of the entire
audio data sample (e.g. an FFT of a short audio file).


OK.

So nothing is changing in the frequency domain.

If nothing changes in the frequency domain for a short audio file, then it
probably isn't a file of naturally-occuring audio signals.

In any case, I use the word convolved rather loosely
here, because quantisation distortion does not obey
superposition.

The audibilty of many forms of distortion can predicted
from spectral analysis of the distortion and comparison
with masking curves.

The first most general rule of thumb is that if it is
100 or more dB down, it is inaudible for sure.

Have there been any tests for realistic jitter spectrums
(the noise on the clock) with realistic audio signals?

Of couse. Check the Banjamin and Gannon JAES paper.

Extrapolating from results using deterministic jitter
with a test tone audio input sounds wholly inadequate to
me, especially given the whole masking effect is
psychoacoustic, and therefore should not be expected to
obey superposition, especially when correlated distortion
is concerned.

I think you're wasting time by worrying about improbable things. Be my
guest, but don't expect me to agree or help you. ;-)


.

Relevant Pages

  • Re: WIMA MKS4 400V Polyester caps
    ... :> of what you can expect with musical signals than a standard 2 tone IM ... and that for almost all audio ... we have to do several filtering and correction background processes, ... IM distortion at higher levels/frequencies, background noise modulation, ...
    (rec.audio.tubes)
  • Re: Scottie Terrierborgs ongoing dread of DBTs
    ... :I did not 'invoke' Arny, ... I was talking about his audio ... his assertions about specs and audibility ... If two items produce identical signals, ...
    (rec.audio.opinion)
  • Re: Better than ABX?
    ... signals. ... Then, using Bill Waslo's "Audio ... A consumer player is certainly not good enough. ... I don' t know anything about the short term stability of consumer ...
    (rec.audio.opinion)
  • Re: BBC2 on UHF38?
    ... If you can see a picture, you can hear the FM sound. ... This used to be generally true when the audio was 6dB below the video. ... Of course reception of IF images of VSB broadcast signals will have no ... twisted together the incoming and outgoing cables and the flylead. ...
    (uk.tech.broadcast)
  • Re: ruthenium condemns ABX
    ... I am aware of some aspects of those physics. ... Generally, in at least line level audio signals, driven from a low ... Switching an audio signal at microphone- or phono cartridge levels is ... I'm not sure if this is comparable to relay switch action. ...
    (rec.audio.opinion)
Loading