[Ardour-Users] Plugin help

[Lamar Owen]

[Sorry for the length]

On Thursday 24 December 2009 07:09:59 am Ralf Mardorf wrote:
> IMO restoration and noise removal is something that is uninteresting for
> audio engineering. I prefer vinyl, but if vinyl is broken, then it's
> broken, there's no way to do any useful restoration with any tool for
> any OS. How should this work? It's impossible. 

If the exact frequency content of the 'pop' is extracted, it is theoretically 
possible to restore what should have been there, at least to a really close 
approximation.  The difficulty lies in determining what a 'pop' is.  And crackle 
is a whole 'nuther issue.

> We should focus on prevention instead of postprocessing.

That's nice in theory, but, when you are doing things like transferring a 
radio station's library off of vinyl and onto hard drive, you have to do 
something.  Or rescuing a little old lady's collection of big band 78's and 
putting them on CD (for her and only her, of course).  Or rescuing your own 
collection of 78's that includes stuff done by now deceased family members.

I've used Audition's click/pop filter before, and wasn't as impressed as I 
would like.  Manually doing a marquee selection in the frequency domain worked 
better than the automatic filter for most of the pops.

Since I have an OS X box, I might investigate the plugins Seablade mentioned, 
if they aren't too expensive.

And I do actually know some folks who prefer the crackle sound.... but the 
typical radio station production/music director isn't one of them (and I do 
some occasional contract production for radio stations, typically batch 
transfers from DAT, reel, 8-track, vinyl, and, believe it or not, HiFi VHS 
videocassette, which was used for a lot of long-form live music recording in 
the early 90's, at least around here).  

I have two SGIs (an O2 and an Indigo2) specifically for the DAT case, since the 
SGI firmware for the Seagate DDS drives can read Audio DAT's, and IRIX has the 
tools to do direct audio extraction painlessly from audio DAT; I've tried the 
linux dat tools, but using them is not quite as painless as using the IRIX 
tools is, even counting having to then transfer the audio files over.  So I 
keep the old IRIX boxen running...

Have a really heavy rim drive broadcast turntable and tonearm combo, with a 
broadcast preamp, for the vinyl, and a selection of styli and cartridges for 
different groove widths, conditions, and the like.  78's really do need a 
different stylus for the best results....

Ok, war story follows.....while an ounce of prevention is worth a pound of 
cure, sometimes you don't have the luxury of prevention.

The oddest noise removal task I've had to do was removing some water heater 
control sequences from a soundtrack that had been printed in 1990. 

Yeah, I'm not making that up; the studio the soundtrack was recorded and mixed 
in had a power issue; when the local electric co-op sent down the control 
codes over the power lines to shut off and turn on water heaters, you'd get the 
high frequency code everywhere; you could hear it in the incandescent lights 
it was so strong of a signal (the filament in the bulb would resonate quite 
well with this signal).  And it got into all the mic wiring, the console's 
power, everything.  You could see it in the 8 foot fluorescent lights as 
waves....

But a guy wanted to use this soundtrack for karaoke-type performances, and 
wanted that annoying noise gone.  Audition's frequency domain marquee 
selection tool saved the day; since the codes were very high frequency (the 
lowest component was above 8kHz, and it ranged up to about 18kHz, with a 
'carrier' around 12.45KHz), multiple rectangle marquee selections of the 
carrier and the modulation 'bits' in the offending signal, and a delete, and 
it's gone.  

And this is a signal that is only ten dB down from the instruments' 
fundamentals, and is 30 to 40 dB louder than the surrounding frequency content 
in the 10 to 14kHz band.  The marquee tool could be considered an automated 
notch filter on steroids...and, honestly, I'm not sure how I could do this in 
Ardour or virtually any other non-destructive editor... but in Audition, with 
the marquee tool, it took all of two minutes to do, once I had located all of 
the occurrences (which just required a listen through, as they are very 
obvious!).

The guy running the studio had tried a number of other techniques, and so had 
I for several years, prior to getting a version of Audition that could do 
that.  With the other techniques, it was very difficult to not harm the other 
frequency content (a notch wide enough to get the modulated bits was wide 
enough to hurt things).  But with very narrow FFT bins, and judicious use of 
the marquee tool, you couldn't hear any degradation in the highs, even on the 
high hat hit that got clobbered by one of the code sequences.  

If it had just been a single track of a multitrack session it would have been 
one thing, but the only format we had the soundtrack in was the fully mixed 
master print (it was recorded on eight track reel, the multitrack tape had 
been lost or reused years before, and the mix was printed in stereo on HiFi 
VHS videocassette at SP).  So copy/paste of another hit wasn't an option.  And 
it wouldn't have helped anyway, since the signal was present in all channels 
(inverse correlation of the channels would have been able to cancel it out, 
but, again, we didn't have the multitrack to work from), and it was present in 
other areas which couldn't easily be copy/pasted from some other area of the 
songs.

That studio tried all kinds of power filtering to get rid of the signal (since 
it got into all their stuff!), and nothing worked.  I mean, you could hear it 
in the incandescent lights, for crying out loud!  Then one day about ten years 
ago, it just disappeared. I guess the co-op went with another system, or got a 
better filter, or something.

And then this other guy wanted to use the soundtrack for his live vocals... 
and, honestly, the noise is pretty jarring when used in a soundtrack/karaoke 
setting with a male voice, not so much in the full instrument/vocals mix the 
original group released, as their vocals (which included a woman singing a 
high part) masked it pretty well, and it was on cassette anyway, and with 
cassette high-end roll-off being what it was back then... 

So I keep Audition in my WinXP virtual machine on VMware just in case I need 
to do that kind of frequency domain surgery again; at least until I find 
something in Linux that can duplicate that functionality.