I have tried this on print through, and also for other uses, such as
removing dialog from a full mix when I have a separate dialog stem.  It is
fraught with problems.  As John Chester mentioned, there are time domain
(length and therefore phase are different) and frequency domain (print
through is muffled compared to desired signal) issues.  Phase (not just due
to the timing differences, but due to the frequency response differences)
is also a deal breaker.  Using an EQ to match the frequency response
changes the phase response so the two will no longer cancel.  I suppose a
linear phase EQ might help in that regard.

I have gotten this technique to sorta work for a few seconds, but to make
it work over the length of a song or a reel of film is hopelessly tedious.
Spectral repair software such as Retouch or Izotope, or if it is voice
only, simple editing in tape hiss over the print through sections, are the
only methods I've had any success with.

I suppose combining frequency, phase, and waveform matching processes might
make this technique work someday.  We've done some pretty amazing things
with wow removal using waveform correlation techniques, though phase shift
is a huge problem there as well.

I haven't tried any de-reverb plug-ins yet, but from what I've read, they
work on a similar principle to de-noise algorithms (multi-band downward
expansion) and have similar (watery) artifacts when over-used.

-- Ellis

On Fri, Oct 25, 2013 at 8:37 AM, Carl Pultz <[log in to unmask]> wrote:

> ... Flip one out of phase and mute it during the desired audio. During
> moments of print-through,
> fade up the other track. It hopefully will cancel the echo, but only
> minimally effect the uncorrelated tape hiss. You might filter the second
> track to concentrate its effect on the particular band(s) causing trouble,
> in order to minimize weird coloration of the hiss or ambience spectrum.