On Sat, 04 Jun 2005 19:22:46 -0400, Joe Kesselman
wrote:

Thanks for the feedback, Edi.

Which one is it?

Well, it's labelled as "Cakewalk Audio Restoration". It's a Direct-X
plug-in, which I wasn't aware of when I first started using Pyro. I know
nothing about what engine it's using behind the UI.
-- It seems to me that Cakewalk -- that is, its makers Twelve Tone --
often bundle with, or sell a known product. Many of denoising software
is made the similar way and have similar controls. Mostly, they
require a noise print, which is either made manually by picking a
portion without useful modulation (at tapes and records, where there
are pauses and lead-ins, it's easy -- but picking a quiet portion
amidst a recording, one must be very aware what's useful and what
isn't). Then there are controls for setting the desired frequency area
and the amount of reduction. Yes, many of them have a "noise only"
switch for checking. Some have additional commands related to FFT
windowing, smoothing, overlapping, re-equalization etc. A very few
tools aren't FFT but are splitting the bands the other way.

The controls are on about the same order of complexity as Soundsoap's
basic mode -- ie, a train/run switch, and sliders for "reduction" and
"sensitivity". The user's clues boil down to "try adjusting with
reduction until you like it; then, if you want, try adjusting
sensitivity. Overdoing either may remove some of the music." (I don't
blame them for not trying to explain this to non-techs, but _I'd_ like a
better understanding of what those parameters actually translate to...)

Basically, the signal is split into many bands and each band is then
compressed to values set from the noise-print information or manually.
This induces comb filtering and is really an eerie thing. So the prime
concern is for me, not "how much I am removing the noise" but rather
"how far I will allow for the whole material to be comb-filtered". It
is impotant to me that, after learning in practice the merits and
culprits of these tools, I don't sigh for some extremely expensive
tools (as allegedly "better" because so expensive) but I rather check
to what extent I can go in noise removal by the tools and if there
were alternate solutions if possible. The most prized tools are your
own skills here and results aren't spectacular but yes, better. If you
have a mint 78 RPM record, cleaned well and played back properly
(geometry, the proper speed, equalization curves and stylus!), it is
easy to make a pleasant result. With totally greyed out records, you
are in fact doing a salvage work as far as it goes and the result is
as is. This is the point, not the "magic" of the hardware and
software. A propely digitized material is 70% work done.

And there's a checkbox for click removal -- again with "just fiddle with
it and see if it helps" instructions -- and a Remove Clipping option
which is probably intended to be a corner-rounder but which so for seems
to be non-usable.
Well, "Remove Clipping" is a last defense ditch for a clipped
recording. If the clipping is (very) mild, it could be remedied. But
the "declipper" should not only round up but compress too.
Some denoiser tools will introduce, by design, a sample shift,
sometimes a huge one, thousands of samples. Such results can't be
inverted and mixed in.

At least, not unless you know how many samples that shift is by and can
juggle it back or delay the original... but, yeah, that's why I figured
it might not be worth attempting. Soundsoap does (or at least did) have
a built-in operation which allows monitoring "just the noise";
Cakewalk's tool, as I said, unfortunately doesn't.
The author of Voxengo tools (otherwise a very fine set!) is aware of
the sample shift and he informs you about the amount of it in readme
files. Then he has made a separate plug-in for bringing the shift
back.
http://www.voxengo.com/ -- I think it's the URL.

If you know the amount of shift induced by a plug-in, you can also
download a free plugin from
http://www.analogx.com/
called the "Sample Slide". Check the other plug-ins too; they are good
and they are free to download.
Any artifacts means you've gone too far already. With these tools, you
should listen to changed top ends and changed quietest signals such as
room acoustics, reverb tails etc. They should not sound numb or
swallowed (ie. too numb; they will sound numb to an extent).

That works for me. Many thanks for the sanity-check!

You're welcome and good luck in restoring!

Edi Zubovic, Crikvenica, Croatia