Any plans to do a True Peak limiter?

[distante]

Well that, some countries have legislation regarding the require dBTP maximum that an audio program have to be, in this cases a True Peak Limiter should be good when there's no time to review everything.

Are there any plans to do a Melda TP Limiter?

Cheers!

[sinkmusic]

Isn't it what Multramaximizer is doing ?
http://www.meldaproduction.com/plugins/ ... aMaximizer

[jmg8]

I think this already exists in MDynamicsLimiter. Look at the advanced settings, there is a true hold setting.

[jmg8]

Isn't it what Multramaximizer is doing ?
http://www.meldaproduction.com/plugins/ ... aMaximizer

MUltraMaximizer is a multiband limiter, I think he/she is requesting a broadband limiter.

[distante]

*He is asking

And yes a broadband intersample limiter for dBTP

[vectorwarrior]

My understanding is the true peak limiting is simply limiting with oversampling. Makes sense really, true peaks are just any peaks that can sneak up between the samples, by over sampling you are ensuring that these extra peaks are accounted for.

MDynamicsLimiter is meldas broadband limiter, and like most melda plugins supports an extremely high 16 times over sampling. Should work just fine

It is not the true peak setting, that is a different way to calculate the peak hold value. This is not connected to true peak detection and is a whole different topic. Essentially peak hold 'holds' peaks in the signal detector, causing a different sound when heavily processing using this signal, but is primarily used in conjunction with look ahead to ensure instant attack times, well, technically it moves the attack portion of processing earlier in time than the signal, but I'm going off on tangents now!

[MeldaProduction]

Aaaaah... those true peaks again ...

MDynamicsLimiter, 4x upsampling (or more if needed), a set the ceiling a little lower than you need. If you want say TP at -1dB, then set ceiling to say -1.5dB. If you overdrive the limiter a LOT, you may get say +0.4dB TP, so that would make it -1.1dB.

But quite frankly the whole thing is idiotic... There are no true peaks, someone invented it to sell some stuff yet it cannot even be defined... It depends on the actual filters and you don't know what the user will use! So just set the ceiling a little lower, done... You shouldn't overdrive it anyway.

[frizzbee]

But quite frankly the whole thing is idiotic... There are no true peaks

Its really funny to read in many threads how idiotic everything is, if vojtech doesnt unterstand the thing. And of cause how all the others are amateurs except himself Makes me smile everyday! Praise this genius of comedy

[MeldaProduction]

Then care to enlighten me? I explained why I think what I think, you just criticise . You can start by explaining what true peaks are, please be specific .

[distante]

The thing is that if the client ask for it, I must deliver

http://www.r128audio.com/

- The Maximum Permitted True Peak Level of a programme during production must not exceed -1 dBTP (dB True Peak), measured with a meter compliant with both ITU-R BS.1770 and EBU Tech Doc 3341.....

Anyways, I'm using this now: http://www.kvraudio.com/product/limiter ... ladg-sound

[Tagirijus]

I also used Limiter No6, till I found a more compact tool:

http://www.kvraudio.com/product/maxwell ... vely-audio

Just by the way.


To the true peak thing:

To my understanding True Peak Limiting means to really limit the audio to the peaks, which would occur while transfering digital audio to analoge audio. Isn't digital just some kind of interpolation, while analoge is continuous? So I understand it that way: when 2 or 3 points of this "interpolation-base" will be interpolated to a real waveform, it could be that the peak goes over the points values and generate clipping, while True Peak Limiting somehow knows and avoids it?

I can at least say that I tested non-True Peak Limiting with True Peak Limiting by analyzing the result audios with this tool:

http://r128gain.sourceforge.net/

After extreme True Peak Limiting (-4 dB for testing purpose) of a signal the R128Gain results where really -4 dB. Not with a non-True Peak Limiter. But don't ask me why, since I am not an engineer.


I have to confess though: if Vojtech would prove me somehow that there is just a trick or similar, I would trust him. At least all his plugins really just work extremely well - that's for sure (to me). So this make me think the way that Meldaplugins have to be made by someone who really knows stuff, hehe.

[MeldaProduction]

Thank you Tagirijus!

So : When a D/A convertor tries to reproduce your digital signal into analog, it internally upsamples it, A LOT, I just googled that 128x and more is often used. The goal is to make the output sound like the digital signal, but without oversampling it would cause quite some artifacts. These would actually be mostly higher frequencies imho, not aliasing, but it's hard to say.
So anyway, the converters oversample it, IN DIGITAL DOMAIN, so the output should not exceed the 0dBFS, otherwise it would get clipped. After that the actual conversion to analog happens, so if something was clipped during the oversampling, too bad.

But here's the problem: We do NOT simulate that with true-peaks, definitely not with the R128 system!! If I remember correctly, then the definition of true peaks in R128 requires oversampling to at least 192kHz, that's it. So classic 4x upsampling.


Is this really true peak detection? No it is not!

It just increases the chance that you detect the actual true peaks, nothing more nothing less. And a limiter that actually delivers no true peaks according to R128, may cause even bigger true peaks in reality (I hope they don't do that though ).
If you would really want to detect the true peaks, you'd need quite serious analytical approach, I didn't think much about it, but I think it wouldn't be exactly simple nor CPU efficient. The R128 is a very bad approximation (like everything from R128). I understand that, it's a complex problem, so they wanted to find a way to do it simple.
Why do you think they ordered -1dBTP? That's because they know the method is far from perfect, so just to be sure...


So, you are forced to follow R128 guidelines AND you want to have no true peaks to ensure your audio will sound good. How?

Use a good limiter with 4x (or more) oversampling AND ceil it less than you would. Let's say you want true peaks at -1dB, then set ceiling to -1.5dB. Afterwards just measure it is compliant with R128, just in case, if you didn't push the limiter too far, you will be just fine.


I just tried an experiment with MDynamicsLimiter 4x and passed the most horrific signal I could think off - rectangular bursts of rectangular waveform. I also tried some normal song, overlimited it as much as I could to make it unlistenable . The true peaks were about +0.4dB above the ceiling, but when I tried messing with the frequency of the signal generator I could actually get to +1.4dB anove the ceiling. The chances your audio would contain something so horrific are close to none, but still... But then I checked the suggested Maxwell Smart and I got more than +5dB above the ceiling!!!
But how did I measured it? Using MLoudnessAnalyzer. So I tried the ToneBoosters' one too, it didn't report so high values, but still above +2dBTP for the Maxwell one. MDynamicsLimiter outperformed the Maxwell's quite a lot, but that one is free and as I said I used extreme signals...


Conclusion?

Every true peak meter reports something different depending on the filters, the upsampling ratio etc. So does each D/A convertor. So we cannot even successfuly measure true peaks, not even according to the supersimplified R128, so limiting according to it is a pure virtue. But you can simply use a good limiter (like one of ours ), set 4x upsampling or higher, set the ceiling a little bit lower than you want and you are done. And of course:

1. Do NOT push your limiter too hard!

2. Do NOT be scared of the true peaks, they won't kill you...

[Tagirijus]

I just tried an experiment with MDynamicsLimiter 4x and passed the most horrific signal I could think off - rectangular bursts of rectangular waveform. I also tried some normal song, overlimited it as much as I could to make it unlistenable . The true peaks were about +0.4dB above the ceiling, but when I tried messing with the frequency of the signal generator I could actually get to +1.4dB anove the ceiling. The chances your audio would contain something so horrific are close to none, but still... But then I checked the suggested Maxwell Smart and I got more than +5dB above the ceiling!!!
But how did I measured it? Using MLoudnessAnalyzer. So I tried the ToneBoosters' one too, it didn't report so high values, but still above +2dBTP for the Maxwell one. MDynamicsLimiter outperformed the Maxwell's quite a lot, but that one is free and as I said I used extreme signals...

I have some questions:

I did tests as well.

I analyzed the audio with R128Gain and with MLoudnessAnalyzer. I crushed the signal with MBitFun and set up ouput to +4 db. R128Gain says my exported audio is +1.1 True-Peak MLoudnessAnalyzer says it is +8.5 True Peak.

Question 1: How can there be a difference?


Then I tested:

a) Limiting with MDynamicsLimiter: ceiling -2db and upsampling 8x.
b) Maxwell Ceiling -2db

The results:
( R128Gain ) + a = -1.3db True Peak
( R128Gain ) + b = -2db True Peak
( MLoudnessAnalyzer) + a = -1.5db True Peak
( MLoudnessAnalyzer) + b = -2.00 / -1.99db True Peak

Question 2: How is this possible?


Maxwell introduces a latency of around 3.99 ms here, but MDynamicsLimiter with 8x upsampling 17,41 ms.

Question 3: How can it be, when the technique (according to your post) is limiting with high upsampling, that Maxwell does not consume that much CPU and does not introduce so much latency?


Please do not consider my questions offensive or similar. I just want to learn - maybe I am doing something wrong here? And besides: in the end it's just rather scientific and not really that important, I guess.

[frizzbee]

Maxwell introduces a latency of around 3.99 ms here, but MDynamicsLimiter with 8x upsampling 17,41 ms.

Question 3: How can it be, when the technique (according to your post) is limiting with high upsampling, that Maxwell does not consume that much CPU and does not introduce so much latency?

AFAIK upsampling does not introduce any latency. But the lookahead does.
And I also think you dont need to upsample the whole signal but the critical parts (the ones that indicate there could be an inter sample peak). I mean if there are just 0 as input you dont need to oversample. There will be no inter sample peak
So in my opinion the oversampling in Meldas plugins is a generell one while the maxwells is specified in true peak detection. This could explain the cpu consumption. Also the maxwell does only upsample 4x but you tested MDynamicsLimiter with 8x.

[Tagirijus]

Also the maxwell does only upsample 4x but you tested MDynamicsLimiter with 8x.

But shouldn't be 8x give even better results according to Vojtechs post?

I understand your other points.