Stereo Tool

New method found. Should be very easy to implement and relatively efficient.

THIS FILTER IS WORKING!

It is still very buggy and I hear a lot of cracks and short bursts of really bad distortion in the sound, but the concept seems to work very well!

Not only that, but I think I can prove that the output (when the bugs are resolved, assuming that I can resolve them) is the BEST POSSIBLE APPROXIMATION of the original signal, based on all the present data. This is not an approximation by making estimations from 2 different sides of a clipped region as other applications apparently use - this thing really generates a formula that describes the simplest function that goes through all the known samples - and then it uses that formula to calculate the missing (clipped) samples.

I've run a test with some of the most difficult tracks from the Torture Test (original_wave.flac), clipped at 1/4th of its peak level, and - except for a really really HORRIBLE amount of clipping and static, which really needs to be fixed to make this usable - except for that, it reproduces the original sound perfectly - with apparently NO DISTORTION at all. (I say 'apparently' because it's sometimes hard to determine due to all the static).

Short teaser (first clipped sound, then repaired sound):
http://www.stereotool.com/temp/TEASER.mp3

LoooooL ! omg!

This kind of distortion is not expected! but anyway ofcourse awesome recover! Is this real ?

How about lows and mid-lows?

Lows are problematic - although I think (hope) that that's the same problem that is also causing this weird static.

I did find at least one thing that I can add to improve the reconstruction. So expect an even closer reconstruction tomorrow

U2 - With Or Without You comes out with an insane amount of static which makes it a bit difficult to properly determine how good it is - but I have the impression that - after clipping at 1/4th of the peak level! - it comes out perfectly. The input contains very very badly vibrating 'S' sounds, in the output that's gone!

I tryed SeeDeClip erlier .. and tryed few days ago...

What i 'see' at output, is that more work is on mid-lows and lows, mostly clipped drums.

Then to confirm i try to clip myself some track to get slightly distorted voice and highs, and mids and highs are almost untouched.

Track i found, belive or not original is like that, is almost not touched by SeeDeClip at all !
Maybe because bad tilt in track ?
http://www54.zippyshare.com/view.jsp?lo ... y=21061145

The static is not completely gone yet, but it has been reduced quite a lot. (I still don't know exactly what's causing it, I just added some information about the input signal that I didn't use before).

Teaser #2:
Original: http://www.stereotool.com/temp/original ... PPEDx4.wav
After declipping: http://www.stereotool.com/temp/01_origi ... PPEDx4.wav

Note: The Original is the recording that I used - if you have another declipper that you use, you can feed it through that to see if it gives better (or worse) results. Clipping level is at 1/4th of the maximum level (+/-8191).

I've tried some 'real world' clipped signals, but due to the tilt in what I've tested so far that didn't work very well. I really need to figure out what this tilt is...
-- Edit: Ok, tilt seems to be exactly what 'Stereo: Stereo Image: Phase' does. So that's easy.

Note: I have just re-uploaded the 2nd file after adding a static filter

Sounds that are a lot (too much) louder in the restored version than in the original are reduced now.

This removes most of the static. But I still need to check:
- If I can prevent it from being created in the first place (probably not completely!)
- With or without you: Loud bass disturbs 'S' sounds too much (most other sounds are restored very well!). This is probably very difficult to solve (has something to do with the nature of 'S' sounds).

Static problem is fixed! And that also seems to have improved the filtering itself.

I've re-uploaded the 2nd version, the result is now really spectacular!

Hm, that's something completely different...