Stereo Tool

Yes.. but that's not nearly good knee as it should be. 0 and 18 knee should be big difference.. especially if ratio is inf:1.
Also am testing Peek based which is more flat release with more thresh - which is unlogical.

Peak does indeed get flat - I know.
Will check knee with some other processors, I thought it looked nice and smooth.....

Ok, I see what you mean - I suppose your response also means that the other knees *sound* better?? Because the shape that I have actually makes more sense to me...

The knee in Reaper isn't implemented correctly btw (start is bad, end is for a while *louder* than the level it ends with), the one in Sonitus looks good though.

Try Sonitus.

Ok, approximately the same shape as Sonitus now for 10 dB (at the threshold Sonitus and Reaper are both 2.3 dB lower with 10 dB knee, Stereo Tool is now 2.4 dB lower), slightly (but only very slightly) less smooth at very low values (3 dB). Above 10 dB I'm getting the same bump that Reaper has, so I'm limiting it at 10 dB now.

BETA017:
Stand alone: http://www.stereotool.com/download/ster ... 04-017.exe
Winamp DSP: http://www.stereotool.com/download/dsp_ ... 04-017.exe
VST: http://www.stereotool.com/download/vst_ ... 04-017.dll

- Fixed knee
- Removed some settings that were no longer functional
- Changed default value for "Dynamic release to 0" slider to 0
- Added "RRMSS" mode (Root-of-Root Means Square-of-Square), behaves more like peak based but it's not.
- Adjusted RMS calculation (there's RMS where a sinewave with peak level 0 dB is said to be 0 dB, and one where a square wave at 0 dB is said to be 0 dB, the difference is 3 dB; my calculation differed from what other compressors use).

Plan for tomorrow: Peak-level based compression smoothing, needed for release behavior. Check if Release behavior, that changes when the threshold is changed when feedback is used, behaves similarly in other processors.
And I want to read/scan through the texts that gpagliaroli pasted (already checked part of it).
- GUI idea

Peak method with knee is now at a usable CPU load, where "usable" means I can at least change things in the GUI. Doing other things on the computer while listening to audio is painfully slow / pointless. On the bright side, that was already pretty much the case with loudness enabled...

All other detection methods give stuttering audio and only minimal timeslices going to other processes. Takes 30-45 seconds to simply end the winamp process.

Without loudness enabled, I can use Smart RMS, although the load goes to 65-70.

Confirmed, in other processors the behavior is always the same, also with feedback. So I'm probably doing something at the wrong place (I'm guessing knee & ratio).

Edit: Ok, I'm starting to understand what's happening.

My compressor determines the TARGET amplification level and then adjusts towards that level. Since the target amplification level cannot be higher than 1, when the feedback level is very soft and an adjustment to a level higher than 1 would actually be needed to reach the threshold, the amount of adjustment is less because the difference in level is small.

Other compressors seem to just calculate the difference in level, without taking things like knee and maximum amplification etc. into account, then adjust towards that level. With hard knee, that means that it can look like a straight line that abruptly stops. But when you use soft knee that's not the case, which means that also the knee is done afterwards.

I'm guessing that moving the knee behavior from measurement to release will give a better and smoother sound. But of course that's not that easy...

To add to the confusion, I've tested different compressors and they all do different things.
Sonitus has the same release behavior as Stereo Tool - release slows down near the threshold.

I have a new idea that may or may not sound good, will implement and post it here (switchable between old and new behavior). This new idea will also affect attack - that's why I'm not too sure if it's good or not.

In my view, at least the basic theory of compressors, the attack and release behavior should not change depending on the knee, you should always keep the same. Although each manufacturer adds some uniqueness to their computers.

For compressors there is no "Target", this is for the AGC's.

Compressors in which there is a transfer curve which determines the output and input Fucion where the maximum gain is equal to 1 (ratio 1:1).

The transfer curve starts at a ratio of 1:1 until it is amended its slope at the point where you set the Threshold value and thereafter change the ratio to the value set.

To soften the abrupt change of gain, the knee is created, which makes for a transition in the gain change in a given window (knee value in dB, are few dB before and after the Threshold).

During this transition ratio value changes to be applied, depending on the selected curve for the knee, which in turn makes before it starts to compress the Threshold value.

Outside the window of the knee applies the Ratio set and before the value 1:1.

As you are all signs before Threshold least knee, no shifts, passed at the same level in the next.

The signal that falls within the knee, will decrease and the resulting gain value according to the transfer curve.

And after knee window is reduced by the value set by the Ratio.

In short, the transfer curve is created based on the Threshold, Ratio and Knee, used to calculate the gain to apply to the input signal, and this gain has values ​​between 0-1.

This gain value obtained (which is actually attenuation), is multiplied by another factor which is the attack or release, whichever applies.

After applying the necessary reduction in function of the transfer curve, set gain is applied.

I hope I can help the understanding of the problem and not confuse more ...

_________________
by GAP
"Less is More" (Bob Katz)