Declipping/Peak Limit Restoration

dallasjustice

Member Sponsor
Apr 12, 2011
2,090
0
0
Dallas, Texas
#1
I want to know whether this is something which can be done with any success. I tried a vst plugin called perfect declipper and it sucked. I've heard a few demos of the izoptope rx declipper which sounds much better to me. Has anyone figured out the best way to "restore" those lost dynamic peaks due to the loudness wars?

The problem is that there's some great music which is unlistenable due to crushed dynamics. My theory about fixing them is best explained by Bob Dylan.
"When you've got nothing, you've got nothing to lose."

Michael.
 
Last edited:
Apr 3, 2010
16,022
0
0
Seattle, WA
#2
I have not tested any of this but did look at the web page for perfect declipper and it goes way outside of what is possible to reconstruct the waveform. I think the right solution only attempts to restore small amount of clipping. More than that may be akin to guessing where the stock market will go tomorrow :).

A side-effect of any such fix is that the levels will have to be pulled down to create enough headroom for reconstructed waveform. As such, the comparison should be done with the levels re-adjusted to be the same as the original in analog domain.
 

dallasjustice

Member Sponsor
Apr 12, 2011
2,090
0
0
Dallas, Texas
#3
Yes. It seems like I could expect to see a -6db or so reduction. I noticed harman has the clari-fi declipper. Is that just for car stereos? Why is it pateneted?

I have not tested any of this but did look at the web page for perfect declipper and it goes way outside of what is possible to reconstruct the waveform. I think the right solution only attempts to restore small amount of clipping. More than that may be akin to guessing where the stock market will go tomorrow :).

A side-effect of any such fix is that the levels will have to be pulled down to create enough headroom for reconstructed waveform. As such, the comparison should be done with the levels re-adjusted to be the same as the original in analog domain.
 
Mar 20, 2011
58
0
6
North Texas
#4
A side-effect of any such fix is that the levels will have to be pulled down to create enough headroom for reconstructed waveform.
But when you pull the level down to create headroom you lose resolution and add quantization noise to the entire track because you have cut off the least significant bits of every sample.
 

dallasjustice

Member Sponsor
Apr 12, 2011
2,090
0
0
Dallas, Texas
#5
But when you pull the level down to create headroom you lose resolution and add quantization noise to the entire track because you have cut off the least significant bits of every sample.
If you didn't dither, that would be true. I am talking about tracks with very limited DR. So I'm not too concerned about "losing" any bits. They aren't being used anyway.
 
Apr 3, 2010
16,022
0
0
Seattle, WA
#8
But when you pull the level down to create headroom you lose resolution and add quantization noise to the entire track because you have cut off the least significant bits of every sample.
Presumably the application here is on compressed soundtracks so losing resolution is the least of our worries :).

Clipping is far more of a problem than resolution loss anyway.
 
Apr 3, 2010
16,022
0
0
Seattle, WA
#11
Hi Bruce. If the clipped section is highly correlated to what is before clipping, then reconstruction can be pretty good. Here is an extreme example done manually in photography:

Original:


Reconstructed:


In this case the background was predictable so it was expanded. And the wing tip was also so and same treatment has gone there.

Of course the above is the work of a human. Computer version is harder although Photoshop's delete tool works reasonably well in recreating what is not there.