Stereo Tool

well .. i can only say .. this is fantastic!

What is left now is how to determine which samples are clipped and which are not. For these samples it's easy (I clipped everything at the same level). That's also what you get when you directly rip a CD. But in MP3s things may look very different.

Bojcha: About Nelly Furtado track: This thing has 'Tilt' again...

Anyway, I tried the following:
1. Boost the volume by 50% (this introduces a lot of extra distortion)
2. Then pass it through Stereo Tool's declipping filter

Result seems to sound slightly better than the original this way. So if I manage to get the distortion detection to function properly (handle tilt) it should really improve.

I've also tried Katy Perry - Firework, but I don't even see any clipping there Still it sounds horrible.

That's what i say.. Nelly Furtado track, .. SeeDeClip also do nothing with that track!

Try Laddy Gaga - Alejandro .. There is one "drum" constantly in clip... what will filter do there... ?

What's important is to have "Tilt" aka "Phase Difference" on input.
Why?
If i play Some track on PC1 and that goes to mixer and master from mixer goes to another PC with ST.. Big chances is that "Tilt" is changed!
So, on input is easy to adjust that "tilt" with some test tone like 50Hz square, using "input scope"
That way we are sure that input signal has no "tilt" and "de-clip" filter will do better clip analyze.

Hm, lacking a better idea I think I'll go for:
- A suddenly appearing and disappearing relatively flat area at a high amplitude.
I'm going to treat any pixels that match that description as 'clipped'. (Correcting tilt might be a better idea, but so far it seems that the tilt changes even during the same track... Either that or I'm doing something wrong - like using the wrong 'tilt' algorithm).

Idea #2: Add a 'did I actually improve anything?' filter - if something was seen as clipping, and fixing it did not improve the audio (get rid of clipping noise), revert to the original signal.

This is not magic !

You need two envelope follower circuits with different attack times to detect the transients of the input signal and two envelope follower circuits with different release times to detect the sustain of the input signal.

The difference between both envelope follower circuits (one for transients and one for sustain) generates two control signals (one for transients and one for sustain).

Then you have to clone the input signal and multiply one instance with the control signal for transients and the other instance with the control signal for sustain.
This way you split the original input signal in two different signals.
One signal contains only transients and the other only sustain.

So, finally you can re-mix these two signals.
For a dynamic (upward-) expansion you add more gain to the transient signal.

That's it.

By the way:
Compare the dynamic range of the original (first 60 seconds) and the processed version (e.g. with TT DR VST meters or the offline tool - important: you have to split it and normalize both signals to approx. 0 dB) !
This kind of dynamic (upward-) expansion processing proofs that such a (green !) signal sounds much better (more vivid and punchier) than a traditional compressed and peak limited signal can.

@Bojcha, Michi95:

About what I hear in this 'Magic' example: Basically it's just when a loud sound kicks in, the start is boosted a bit. I'm not so sure though if I like the 'after processing' version better than the 'before processing' - yes, it has more punch, but it also sounds a bit unnatural to me (somehow it sounds a bit like MP3Pro compressed audio edit: No, it sounds like recording something on a cassette tape without Dolby, then playing it back with Dolby S. That gives EXACTLY this type of effect! On compressed audio, things often improve if you do this though.). I've compared the original and after processing versions on 2 sides on my headphone, and if I do it that way I really prefer the original. Which is odd, because normally (uncompressed vs. compressed) I usually prefer the uncompressed version. I think this tells me that something is wrong with the 'after' sound.

Looking at the waveforms I think I understand *why* it sounds unnatural: Just every loud peak gets a boost. Regardless of whether it needed to have one in the original or not. While it seems better at a quick glance (at first I also thought it was an improvement), I think it gets tyring to listen to after a while. And I REALLY doubt if it's possible to find good settings for *all* tracks.


Having said all this, I do think that a technique like this - if it works properly - could be very useful. Michi's description did give me a new idea - give me some time to think it over...


Edit: I do think I know now what is wrong with this. Say you take a dance track with loud beep tones.

If you would process that like this, the start of each beep would be made louder - the level is constant so the processing would determine that it is already strongly compressed and attempt to give it a 'punch'. Which will sound just horrible.

Still thinking...

Hm, question about this 'magic' wav file: What are we trying to solve here actually?

Compression? No
Clipping? No

So it's apparently just a matter of boosting dynamic range.

So we take a natural signal, and boost its dynamic range. Hm, not sure if that's really a good idea?

I did not think the "magic.wav" is fantastic.

So, for me fantactis is http://www.stereotool.com/temp/01_origi ... PPEDx4.wav

I already say my opinion about "magic.wav" on their forum! Not here

Aha, but who are you in that forum ?
Just to be sure:
What have you done with the original downloaded WAV file ?
Ok, for sure you have split it in two files (60 s unprocessed and 60 s processed).
But then ?
Have you compensated the peak level difference ?
The original part in the download version is already processed by negative gain to get the necessary headroom for the upward expansion.
So it is really difficult to compare both parts mixed on 2 sides on a headphone.
I guess you prefer the unprocessed version, because it sounds less aggressive based on the loudness difference.
But you have to keep in mind that in reality (normalized to 0 dB) it is the other way round and the unprocessed part is much louder and aggressive than the processed (dynamically expanded) version.

Though I have to admit that I agree: there is a little bit too much expansion.
But if you can adjust the degree of expansion, IMO you can create a version that sounds less artificial.
And important in this case:
It is a live recording with relative constant volume based on the mixing and mastering (compression, etc.).
But if you do it with music from a studio production with much more macro (program) dynamic then it changes almost everything.
IMO (using less agressive expansion settings) you can create a less fatiguing music (compared to a heavy compressed, peak limited master).