[linux-audio-user] Mastering; Rezound, fir filter, digital clipping

[Oyvind Hammer]

Hi Ron, let me try a little on this one.

> I guess the process of copying audio and data from CD
> doesn't introduce any opportunity to compromise the
> sonic quality of the source. Am I right or wrong?

Right, if it's done digitally and correctly.


> Is what I am seeing clipping

This is almost a philosophical question :-)
Not knowing Rezound, I assume that it
simply checks whether the sample values are just
about to max out. For 16 bits, you have a range of
-32768 to +32767, so the program might blink the red
light for samples below -30000 or above +30000, for
example. Rezound can not actually know if the signal
has clipped - maybe it's recorded beautifully at full
dynamic range.

Of course, you can look at the waveform to see if it's
flat on the peaks, but maybe it was meant to be that
way - there are many strange musical aesthetics out
there - the computer can not know.

So to answer your question: The red light does not
technically show clipping within the digital domain,
but I would say that it is a strong indication that
you did have clipping somewhere in your analogue
path way back then.


> Perhaps someone could provide a technical explanation
> of clipping or a link to a definition.

Hm, this is really quite simple, something like
limiting of the signal due to the amplitude
exceeding the dynamic range.
If your original, full-scale samples x are clipped
in a system with a range of -1 to +1, the clipped
signal y will be

y=-1 for x<-1
y=x for -1<=x<=1
y=1 for x>1


> What tools do you use for eliminating clipping that
> already exists in a source? I don't care at all about
> preventing the problem.

Clipping means information loss, and you can not in
general reconstruct the signal. There are all kinds
of clever tricks to try to round off clipping, but
they will always be fudges and not total cures.


> For the moment I am using the Rezound Arbitrary Fir
> Filter to identify the hz where the clips occur

?? clipping will affect all frequencies. Of course,
depending upon your signal, the audible effect of
this will be more serious on some frequencies.
In general, clipping will add energy on high
frequencies (for a pure sinusoid it will add
harmonics).


> performing a decibal cut on the problem range.

Yes, it's not a bad idea to bandstop-filter
bands which sound particularly nasty.


> Reguardless of how detailed I get, there's
> an audible consequence to eliminating the clips.

Yep!


> What's interesting about this specific set of clips is
> that they are mostly inaudible. The clipping occurs
> around 10kHz -> 15kHz and are almost all within the
> high hat.

This is a little strange. Althoug clipping generally
adds high-frequency content (and can be dampened a
little by a low-pass filter, effectively rounding off
those nasty sharp bends in the curve), I would expect
some goo below 10 kHz. Maybe it's masked by your
music.


> *how Rezound is configured to conclude that there are
> clips

See above

> *should measuring for these types of problems be user
> configurable or does a technical specification define
> when a clip occurs

?

> *can engineers safely ignore inaudible clips and tell
> their clients ... not to worry

Ha ha, they do all the time :-)


> Is a Fir filter a good tool for addressing the problem
> of digital clips or is there something better?

There are special, "intelligent" tools out there
(can anyone help us with names?), but no miracle cure.


> Are there alternative Fir filter algorithms that
> produce better results than the one being used in
> Rezound? I haven't a clue what Rezound uses.

Me neither, and there are all kinds of fir filters,
but I doubt if any would be better at solving your
problem.


Dr. Oyvind Hammer
Dinosaur researcher etc.
University of Oslo