MLimiterX - Peaks above -0.5db (set ceiling) - Why?

[TommyC81]

Hi there, a quick one; I notice that for some particularly demanding pieces of music, and also when changing presets in synths etc., peaks above the set ceiling (-0.5db in the default preset) makes its way through the limiter, sometimes as much as +2db.

This is using the default preset (which has a look-ahead of 0.5ms).

I admittedly thought that nothing above the set ceiling would get past the limiter during any circumstances, but this is not the case. Is there any specific setting I should be looking for to achieve a proper "brickwall" (i.e. no true peak exceeding the set ceiling).

I'm using Reaper and using the default peak metering, but also have the same indication in MLoudnessAnalyzer (True-peak).

I gather I can simply increase the oversampling (seems to work better), but am curious if there's anything else that I should be aware of?

[Hexspa]

I've noticed the same and, off the top of my head, oversampling is the solution; could be mistaken.

[MeldaProduction]

When the plugins are loading (or doing other time consuming stuff), they are temporarily bypassed, to avoid CPU overloads. That's probably where this comes from.

[TommyC81]

When the plugins are loading (or doing other time consuming stuff), they are temporarily bypassed, to avoid CPU overloads. That's probably where this comes from.

Unfortunately, it happens when just playing demanding music without any switching ("Intelligent sleep on silence" does not change the outcome).

I've attached some screenshots and also a Reaper test project if it helps - just press play. The project file should cause true peak overshoots to various degrees between the plays.

A random testing for me (as per screenshots) gives:
MLimiterX: +0.4db (target -0.5db)
LoudMax: -0.44db (target -0.5db, this is within the described technical limitations of the plugin)

The project file is based on:
Reaper 6.59
MComplete v16.00d

It also has LoudMax (v1.41) in it for comparison (it is free) - this plugin works as expected. Can be found here, along with technical explanations: https://loudmax.blogspot.com/

[Hexspa]

MLimiterX has a clipper in it right? Do I need to put a clipper in front of it to get into young person's loudness? Would MSaturator work for clipping purposes? Anybody use a clipper here?

[MeldaProduction]

Unfortunately it doesn't seem to do that here . Could you make a screen video? There must be something that temporarily bypasses the plugin.

[TommyC81]

Unfortunately it doesn't seem to do that here . Could you make a screen video? There must be something that temporarily bypasses the plugin.

For sure, uploaded a short test screen capture to YouTube here (no sound):

I've repeated this many times, with and without "smart sleep" and with/without the plugin going into "Sleeping" between plays - I get the exact same result every run with MLimiterX: +0.04db on left and +0.12db on right.

If I change to 48kHz (rather than my usual 44.1kHz), i get slightly better results with less overshoot.

Edit: To confirm that this is done with v16.00e, no change from previous version (v16.00d).

[operator]

I didn´t know that MLimiterX should be a "True-Peak Limiter". If it should be, then yes, I get also true-peak-overshoots. But the Peak-value in the Loudnessmeter is the same as the Ceilling --> like it is supposed to be.

But to be honest... I don´t mind. I have also another (popular) Limiter which behaves the very same way or even worse . We have the Loudnessmeter to dial that back --> just lower the ceilling until you get your desired endresult.

Another one I have, Ozone 5 Limiter has a True-Peak option to enable.

So maybe having such a True-Peak option to enable would be nice, but shouldn´t be the standard (in my opinion)

[ystmrk]

I have read the "True Hold" section of this document.
https://www.meldaproduction.com/downloa ... _intro.pdf

p.s. I personally have not encountered any trouble so far.

[operator]

[...]

Maybe check again with MLoudnessmeter and look for "True-Peak" readout...

True Peak and True Hold is something different I think. True Hold works on the "internally sidechained not audible" detector signal/envelope. True Peak on the other hand seems to be something like a "inter-sample calculation thing" on the audible signal aka. Output (I m not a Dev, so sorry if I may chose the wrong words or am just simply wrong)

And I am looking forward to be corrected... I am eager to learn.

[TommyC81]

I didn´t know that MLimiterX should be a "True-Peak Limiter". If it should be, then yes, I get also true-peak-overshoots. But the Peak-value in the Loudnessmeter is the same as the Ceilling --> like it is supposed to be.

But to be honest... I don´t mind. I have also another (popular) Limiter which behaves the very same way or even worse . We have the Loudnessmeter to dial that back --> just lower the ceilling until you get your desired endresult.

Another one I have, Ozone 5 Limiter has a True-Peak option to enable.

So maybe having such a True-Peak option to enable would be nice, but shouldn´t be the standard (in my opinion)

I've kinda assumed that, with the extensive functionality of the Melda Plugins, True Peak was considered and accounted for in MLimiterX. But maybe that is not the case? It doesn't seem to be, hence my initial post/question.

Chances are that is the answer: MLimiterX doesn't account for True Peak.

If that is the case, then it'd be desirable to have an option for that in a future update.

[MeldaProduction]

Well, from the video I can only see that the true peaks are overs, not the actual peaks! If you want to fix true peaks, set MLimiterX to say 4x oversampling (or more). I have described it several times, but - true peak metering is NOT what people think it is - it's described as measuring peaks inbetween samples, but there's quite no solution for that (or I'm not aware of any). So it is done by oversampling the actual meter 4x, that's all. So to limit for that, you also need 4x oversampling.

[vectorwarrior]

I always use limiterX with oversampling for this reason. Noticed it in reaper myself thanks to the new stats on rendering.

[TommyC81]

Well, from the video I can only see that the true peaks are overs, not the actual peaks! If you want to fix true peaks, set MLimiterX to say 4x oversampling (or more). I have described it several times, but - true peak metering is NOT what people think it is - it's described as measuring peaks inbetween samples, but there's quite no solution for that (or I'm not aware of any). So it is done by oversampling the actual meter 4x, that's all. So to limit for that, you also need 4x oversampling.

Thank you! I've educated myself, and this is the way. Much appreciated!

I found this article from iZotope from 2015 that helped me wrap my head around it, maybe useful for someone else as well: https://techblog.izotope.com/2015/08/24 ... detection/