Does anyone know of a way to correct a non-linear phase shift produced by noise removal with an IIR filter. I assume it's IIR because if it were FIR, I'd think the creator should have taken the trouble to make it linear-phase. I can tell that there is an induced phase discrepency as a result of the filtering because of the added reverb in the filtered sound and because a phase analysis shows that different frequencies are out of sync. I would appreciate any advice on correcting this issue. Ideally if a plugin or application could track all the phase shifts at all frequencies through the length of the entire file, invert them, and mix them, I think the problem would be fixed, but I have no idea how to do that or if there's even software that can do such a thing.

- Blue Streak

Assuming the "filter" to be really an IIR filter:

Here is one option:
- apply the filter
- reverse the time (so the sample will be played backwards)
- apply the filter once more
- reverse the time again
The filter is applied twice now, but any phase shift is 100% cancelled.


Sebi

I suppose I should point out that many noise-reduction operations are non-linear, and as such, there may be more to "inverting" the filter than simply reversing the phase, in fact the meaning of "phase" may be a bit, err, unusual.

Okay, I tried applying the filter by taking the same noise profile and applying it to the same file reversed with the same settings, but when I looked at the phase analysis (In Adobe Audition) it only looked worse. "S" sounds in particular look ugly and show a whole bunch of concentric rings about the axis. However, I've just had an idea, I'm going to try inverting the waveform and then running it back through the filter to see what happens. By inverting the waveform, I'm hoping the amplitude reduction will shift the phase in the opposite direction this time because the sample sign is reversed!

- Blue Streak

Alright.....I don't get it.....it appears that the places where I reduced the noise inverted and normally has an even larger phase shift. Same thing when it was both reversed and inverted. It doesn't seem to affect the voices too much because most of those frequencies don't get phase shifted when the signal is particularly strong, but it's pretty obvious on the analysis graphs. This doesn't make any sense to me.

- Blue Streak