Stereo Tool

Looking at these images, I'm starting to see that in the 018/KneeOnTarget=OFF and 020 implementation, a MUCH bigger knee value (current maximum is 10 dB) would help a lot. And it will also give some extra compression. So maybe I shouldn't give up on it too soon...

In the graphs we can see the errors you mention.

I think in principle, is to plot the way as you have done in order to find a more static where are the miscalculations and also do not think we are far from reaching something that works properly.

Once you have found that the behavior does not make mistakes, you can begin to improve (if deemed appropriate) behaviors, varying the waveform, softened by the Threshold, etc.. and make it more "musical ear"

I think that would be good also plotted vs. Input Output, for a range of input values ​​grow linearly, to verify that the transfer curve and knee are being correctly calculated for different ratios.

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

Point is, they aren't really errors - they are basically different solutions for the same problem, and I see the same things happen in other compressors.

So the question is not so much what the 'correct' behavior is - they are all correct in a way - but which solution sounds best.

Have been visiting an event today and will do the same tomorrow - so no updates for now. I'll do a few more listening tests tonight.

Edit: I really can't say I hear any difference between these modes. Probably because my ears aren't trained yet for compression effects. So I'll have to rely on feedback from others.

When I enable 'Difference' as Bojcha suggested the differences (between "Target on Knee" on and off) are very clear, but I just don't hear them in the processed audio. Sometimes I think I've heard something, but it could be a placebo effect (I *know* I'm changing the setting).


Edit: Just looking at waveforms then.


020 vs 018. As you can see 020 looks much flatter - 018 will probably sound much better (and that's confirmed by feedback I got from users). So I can probably remove the 020 code and continue with 018. 020 also comes up a bit faster - but that's only part of the difference, the shape is really different.

I have to state, up front, that this discussion is a bit beyond me because the last time I was really serious about trying to study this kind of stuff was 24 years ago and I have forgotten a lot of stuff, but I do want to remind you of something.

I've noticed that "difference" is not really all that useful to me in many cases. The ears compensate for things to some extent, and as we learned from the "Aysmmetric Bass" issue, some people hear things that others do not.

Also, ultimately what you did with the spike protection in the AGC was beneficial, but quite a few people felt that that loud burst in "The Pretender" was bass, but it was really the forceful start of the "wh" sound and the attack phase of the percussion and guitar all starting at the same time, thus it was primarily high mid-range, not bass. Clipping in multiband around that area helped, and then the spike protection helped some more.

In other words, keep in mind that sounds have components across larger portions of the frequency spectrum than just their fundamentals. Things that people may attribute to a particular frequency range may or may not be where the problem lies.

I just realized something.

In the implementation of 018 with "Knee on Target" ON, attacks will start at some (high) speed, then slow down when they get closer to the target level.

In 018 with "Knee on Target" OFF, and in 020, if you're not using Knee the attack will start a LOT faster, and go a bit slower after that (because it's already closer to the target level). That's probably bad, because it makes it very hard to find a decent Attack speed. In fact, when the audio level is closer to the Threshold, Attack will get more and more aggressive at the start. --> No this is not true, in fact Knee would help smooth the start of Attacks closer to the Threshold, which might actually be a good thing.

So that's another reason why 020 and 018 with "Knee on Target" OFF would be bad, at least without using Knee. With Knee I'm still not 100% sure.

I did some more tests tonight and I can confirm that NONE of the other compressors I tested behave like "Knee on Target" is OFF. Which means that I can remove 2 of the 4 variants in 018 and the complete version 020.

The dB style in 018 seems to be worse than normal style, so I'm removing that as well.

Also added a new slider to control the bass delay behavior, because of complaints about distortion. Existing presets may need to be sligtly modified, by default the 'delay bass clipping' effect will be reduced by half.


Edit: For that last slider, it probably makes much more sense to use a 'relative sensitivity'. Will do that in the next version.

Beta 021:
Stand alone: http://www.stereotool.com/download/ster ... 04-021.exe
Winamp DSP: http://www.stereotool.com/download/dsp_ ... 04-021.exe
VST: http://www.stereotool.com/download/vst_ ... 04-021.dll

- Chose 1 implementation for attack/release/knee behavior
- Added slider to clipper to reduce distortion caused by delayed bass clipping (Advanced Clipper -> Distortion -> Delayed bass clipping Strength). Not working properly yet - it helps but not enough. Working on it now.

Some good progress today:
- Improved performance of compressor (still not good enough, but definitely improved)
- Added better PEAK based mode, with smooth peaks very similar to smooth RMS
- Improved bass clipper delay
- Added extra bass clipper setting to remove some more distortion without significantly reducing the bass level (I hope).

Beta 022:
Stand alone: http://www.stereotool.com/download/ster ... 04-022.exe
Winamp DSP: http://www.stereotool.com/download/dsp_ ... 04-022.exe
VST: http://www.stereotool.com/download/vst_ ... 04-022.dll

- Added Smart Peak mode to compressor, use this instead of Peak for good release behavior!
- Optimized compressor code (not finished)
- Improved 'Delay bass clipper' behavior
- Added bass clipping level for whole bass region - unfortunately this seems to work very sharply so it's causing artifacts again. Need to make sure first if this filter helps, if it does I'll try to make it less steep.