🚀 go-pugleaf

In article <1993Jan23.074015.13681@news.columbia.edu> gmw1@cunixa.cc.columbia.edu (Gabe M Wiener) writes:
>In article <1993Jan22.155947.4192@wuecl.wustl.edu> wb1@cec1.wustl.edu (Bill Bartman) writes:
>>I am looking for a program for PC which can analyze waveforms and
>>remove noise, clicks, and pops.  What I am planning on doing
>>is digitizing old analog recordings, "cleaning" them up, and then
>>recording onto DAT otr the like.  I have a great stereo, PC, and
>>sound-card.  Now all I need is some software which can manipulate
>>the file on my computer.  Does anyone know of such a program or
>>know of anyone who may know of such a program.
>
>I'd suggest you forget it unless you have five figures to spend.

Actually, it's not all as bleak a picture as you paint it.  With the
DSP boards that drop into a PC, it's possible to do a lot of such
stuff in real time, and a lot more if you can keep the entire piece
on mass storage and run in batch.

I'm currently doing click and pop reduction in realtime, as well as
full inverse Laplace functions in about half realtime, using a 
1989 vintage TI 320 series processor hosted in an aging Microvax 3200
with 32 Mb of core, and S/PDIF interfaces built around a DHV-11 board
and the Crystal Semi chips.  The software is entirely homebrew, and 
is mostly built around a very finely-tuned set of libraries supplied
by TI.

This is definitely not current technology, but I'd say my total cost
for the system was under $5,000 total, not counting the many hours
of scrounging around at government auctions, and of trying to coax
software to run in some semblance of realtime without breaking up
periodically due to I/O bandwidth limitations.  You could probably
do better with a more modern technology, especially if you are willing
to wait a couple of years.
--scott

Secrets of click and pop reduction:  Trigger on the second derivative
of the input signal and not the first derivative.  Don't just cut the
signal out for a constant amount of time: find both the beginning and
ending of the click and use least-squares over a few hundred samples 
to fit them together.  If the click only appears in one channel, use
the signal from the other channel in place of the synthesized curve,
but only if it varies less than a present amount from the curve across
its length.  You definitely need to have a lot of ultrasonic information
to find the leading edges of the clicks; use a cartridge with a very
high point of resonance (I use the Adcom Hypercoil MC), and a sampling
rate of at least 48 Ksamp/sec.