Stereo Tool

And: An idea for 'undoing' limiting with filtering.

I think I have found a reliable way to determine whether sound has been filtered or not, and to restore exactly those sounds that need to be restored. No extreme punch when it shouldn't be there. And this filter is probably very easy to make...

Problem with unstable clipping level is more-or-less solved. Clipping restoration is slightly less effective, but no extra distortion is added anymore - even if I have peaks to 400% in the input signal (ie. I set the clipping level far too low).

I'll add some code to automatically detect which declipping type should be used based on the input signal. I guess I can do that based on a histogram.

And does this work without any phase problems ?
I don't remember who, but someone linked Terry West's free ReLife VST plugin.
And this leads (using HPF filtering ?) to severe phase problems.
Obvious when you push your mono button ( you loose mono compatibility if you use it) .

So, I hope that you find a way without these kind of phase problems.

EDIT
I forgot:

Thank you very much !!!

Michi: What do you think about the improved "American Idiot"?

New peak restore idea: Cleanup + boost should give info about the wanted levels per frequency. Then correct frequency bands (multiband style) to match.

That might actually fix everything...

It is an audible subtle difference (more like the difference between a medium quality D/A and a high end D/A).
No doubt it enhances high frequency definition ( high end ! - ).

Declipping only, but this close to perfection (comparing it with SeeDeClip and Izotope RX Declip).
Comparing this with the UNDO processed version by JesseG it sounds better.
With this track ('American Idiot') it has become more obvious to me than with the magic.WAV that UNDO and its upward expansion concept leads in the wrong direction (at this degree of expansion).
The drum sounds in the UNDO version is much too aggressive (it really hurts in your ears IMO).
But I am sure that UNDO processing could sound better if the boost factor would be lower (50 % of what is used now) or better adjustable for the user.

There is only one point left.
When I look at the waveform of your declipped version there are some spikes that look strange.
Especially the last red spike close to the end.
Isn't this too much (an error) ?
My tests with SeeDeClip had similiar (somehow artificial ?) spikes.

Could you add a setting (adjustable for the user in the final release version) to define the maximal amplification for spikes ?

Sure. Actually it's already there (but not configurable), set to 400%...


By the way: From the SeeDeClip website: Hm. I guess I should write my own. Something like: Stereo Tool uses about 200 lines of code in 3 source files ...

LOL

I don't want to spoil your concentration on this declipping module development, but here is a free transient tool:

Information page:
http://www.braindoc.de/manuals/EnvelopeDesignerEn.html
Download page:
http://www.braindoc.de/english/vstEn.html

Michi:
I'm moving in a totally different direction here as well now. I've created some new processing steps, and on my screen I can now clearly see two graphs, which affect each other. This is for Lady Gaga - Alejandro. I can clearly see in my analysis that peaks in my first graph cause gaps in my second. If I can automate finding the correlation between the two, and restoring the gaps in the second graph, I think I can restore the transients in this track based on what happened in the production process, instead of using other techniques which basically come down to just 'guessing'.

For declipping I didn't think guessing would do - and I think my filter proves the point (remember I wrote it in just 3 days! Eventually the filter is extremely simple!)

Now for restoring spikes that were 'compressed' I want to use a similar technique. And - based on what I see on my screen right now - I think I'm getting close!


See this image:


This is one of the bass kicks of Lady Gaga - Alejandro. At spikes in the total signal (bottom graph, basically that's the extremes of the bass kick), the intrument volume (top graph) drops momentarily. Fixing that will restore the lost punch - and only when it's needed.

What makes things more difficult is that the level of the instrument is not very constant anyway, so it's difficult to determine which drops should be present and which are caused by the spikes at the bottom graph. (It looks like there is a level below which spikes in the bottom graph have no effect on the top graph. That might be part of the solution.) Also, this is one of the clearest examples - most are less visible, or don't really seem to match.

If I manage to restore this, I basically have an 'un-Loudness' filter. And I think I can create a 2nd filter - afterwards - that removes any extra sounds that would be boosted erroneously.

I don't know what kind of tracks you are after for this, but I can tell you a track that is very clip-heavy:

Information Society - "What's on Your Mind (Pure Energy)" (Club Mix)