Different sound levels of the files - How normalization works

I have problems with the different sound levels of the files. In order to solve the problem properly, it is important that I understand how normalization works. At the moment this is what I know about normalization.
During the normalization scan, 3 values are stored. This is very useful because you can change the normalization type without rescanning the files. I am currently using the true peak setting.
So far everything is clear, but now comes the unclear part. Suppose the true peak gives a value of -3 after the scan. What does Mairlist do with this?
With my limited knowledge, I expect Mailist to enter the true peak value in the gain (db) field in the normalization setting, with the aim of amplifying or reducing the volume of the files so that they are all equal.

Often a value has already been entered in the gain field. This does not correspond to the true peak value. I batch set all the gain values to 0 and then ran the normalization again. Again, there is a value in the gain field that does not correspond to the true peak value.

From the above you can see that I do not yet fully understand how normalization works.

Is there someone who can explain to me what Mairlist exactly does with the normalization values and whether there is a relationship with the gain field?
I would also like to know what exactly the loudness scan does?

Hello @Baasb,

welcome to the forum and the community. :slightly_smiling_face:

mAirList offers you two different types of normalization.

  • Peak normalization:
    The audio element is normalized (up or down) to the maximum peak level that is set in the configuration. That means, that the highest (!) peak of the element always reaches this value, and it doesn’t matter if it’s only one peak or many.

    • Full scale (dBfs): Outdated technology. This was the only option in versions prior to v6.2 and is only available for compatibility reasons.
    • True Peak (dBTP): It is a special technology that calculates so-called “intersample peaks” and thus provides you with more precise values. This is particularly advantageous for lossy formats such as mp3.
  • Loudness normalization:
    The audio element is normalized to a target level of its loudness, which is called “R 128” here. Calculating loudness is a bit more complicated and more accurate than RMS (Root Mean Square) values. The EBU R 128 recommendation is based on this loudness calculation.

If you want to have the loudness of your program at the same average level, the second method is recommended.
Well, that depends on the audio material. Modern pop music from the last, let’s say: 20 years, is mastered in a very “flat”, not really dynamic way. It looks like this (2009):

These are the audio levels of this title:

Now let’s have a look at a title from 1988.

The real surprising difference in loudness can be seen in the audio levels:

Well… now let’s have both titles in a playlist, normalized at -1 dbTP:

  • First title: Amplification 0,0 (the True Peak level matches perfectly).
    Playout loudness will be -9,4 LUFS (not “dB”, but this is another discussion).

  • Second title: The loudness must be reduced from + 0,2 to -1 dBTP. The amplification is -1,2 dB, so the playout loudness is:
    -13,2 minus 1,2 = -14,4 LUFS.

In this playlist your listener hears two songs with a different loudness. His listening volume level is adjusted to a loudness level of -9,4 LUFS (usually the volume knob is turned down). Now a title with a loudness of -14,4 LUFS is streamed; 5 LUFS quieter (the volume knob is turned up).
Next song is a bit louder (volume knob is turned down…) - you see the problem?

Now let’s try loudness normalization.
Personally, I use a target loudness of -18 LUFS (temporarily this is allowed in the EBU R 128 s2 documentation).

  • First title: -9,4 minus 8,6 = -18  LUFS.
    DR 5 amplification

  • Second title: -13,2 minus 4,8 = 18 LUFS.
    DR 11 amplification

When you compare the waveforms, you will see the difference using loudness normalization:

The light green waveform is the real (playout) loudness; dark green the part that was original but is now reduced.

The advantage of this method: All titles are played with the same (average) loudness level and the dynamic of the titles remain. So it’s the same with the whole playlist.
If a song has only few dynamics (“flat waveform mastering”), it’s not our problem.

Let me explain you where we might have a misunderstanding:

In the normalization settings of the configuration, you set the target levels.

Each title has a different loudness and peak value. For playout purposes, the value is written in the field “Amplification” (see above) of the element properties. This depends on your normalization method.

When you change the normalization method or the target levels, you need to normalize again. You can use the mass edit function:

:information_source: You don’t need to redo the loudness analysis.

Precisely. :white_check_mark: :sunglasses:

Hope that helps.

My normalisation was set to true peak but there we’re some variations in loudness.
So after reading this last week I thought ok let’s try ULI his 18 luft setting.
and global it’s better but in my broadcast this weekend I had some records that were loud very loud for instance David Bowie Let’s dance every beat of the drum the Vu meter goos high in red.
It looks like the gain is less 3.5 while the most is around 8 a 9
Did I do something wrong?
Can it be that mairlist skip’s songs and that it is still on true peak?

Hi Henk,

thank you for this (very good!) example.

My CD-Rip is an mp2 @ 256 kbps.

It seems that you can see each beat as a peak in this waveform.
In fact, some beats seem to be very present in the foreground, at least that’s my personal listening impression.


David Bowie - Let's Dance (loudness levels)


David Bowie - Let's Dance (amplification)

At the first glance, everything’s fine:
-12,6 LUFS minus 5,4 = -18 LUFS.

Additionally, we have to take a look at + 1,4 dBTP: The negative amplification mus be at least 2,4 dB or below so that the value of -1 dBTP is not exceeded.
Yes, that’s included, everything’s fine.

:information_source: :arrow_right: This is the way how you can check out each element in your database or playlist when you have the impression it could be too loud or “mighty”: The addition of loudness and amplification should sum up to the target value.

A mixdown from the playlist (only this title) shows the results:

It’s exactly what I have expected.

I don’t know, but as good as I know you: No, I don’t think so. :crystal_ball:

Hopefully you don’t use any kind of compression or other DSP (from my personal point of view, there’s no need for it: When you play a CD or a vinyl on your stereo system at home, is there a “compression” or “max. loud” button?).

But here’s a hint: In the last 30 years we have forgotten how good music can sound. We are victims of the so-called “loudness war” (or loudness race) - it’s the same in music production and broadcasting.
Look in this article for “dynamic range”.

The Pleasurize Music Foundation (website offline, see information in this Sound On Sound article) has published a dynamic range meter and I still use the foobar2000 component “foo_dynamic_range”.

This tool has calculated a dynamic range (DR) of 11 for David Bowie’s “Let’s Dance”. Modern pop music often has a DR not higher than 5 or 6: It’s compressed too much.

This might (!) be a reason why we think dynamic titles are louder compared with loudness normalized reduced, compressed music titles: They have less dynamics and, as I said before, we have forgotten how well dynamically produced music can sound compared to the mush of sound that is unfortunately typical nowadays. We have to get used back again to what we have forgotten over the years.

Younger people that have grown up in this loudness war think, this kind of music production and listening is normal. No, it isn’t.

In a digital music production environment we should no longer use VU meters. Did you mean peak level meters instead?

I don’t see a reason why mAirList should do it. Please check the addition / calculation as I have shown you above.

Hey, are you interested in another experiment?
Do you have “Bobby McFerrin - Don’t Worry, Be Happy” in your database?
Check out loudness, true peak and amplification… and have fun! :sunglasses:

1 Like

I mean the led on the airence and the on screen d&r meters (that react faster)
I even had the Orban meter as an more objective meter. And the meter in Mairlist that hasn’t got a scale but if everything stays in the green and David Bowie comes halfway orange my guess is that it louder.

Not sinds I use the Airence before I had first a behering 2000 and after that a Allen Heath xb 14 both had inserts on the output and on the insert was a dbx 166xl.
Back then I provided my own stream with stereotool in stereotool you could say I don’t want the orange in the mairlist scale and even when you don’t normalise you wouldn’t hit the orange.

Now I volunteer for a radio station I use a stream the station picks it up there it goes online ( with some compression)
My only goal is an average sound on my headphone (and speakers) I think the topic starter want the same.
With true peak there was a difference of 2 leds now with r128 on 18 luft the average is (on the Orban meter) -5 and peaks at 0 db and the 0 bd is the complete song it’s not just a peak.

Given that the mAirList-“meters” show something like peak indication:

In many cases, it is. But generally you can’t imply a higher loudness from higher peak level values. Imagine a solo flute with some highly enthusiastic percussion in the background: The loudness will be relatively low, whereas the percussion will yield high peaks but low intensity.

What “is louder” is what you hear louder (or what you read from a dedicated loudness meter, respectively).

1 Like

If you drive you car 90 kmh on the highway it feels like your standing still.
Take the same car to the Stelvio pass (Italy) go from the top with 90kmh down you think you drive 200kmh.
The point where green goes to orange is my limit (Mairlist meter) in the above case Bowie I turned the input volume down underneath the orange and the next record up again and that is the thing way I use normalisation in the first place.
And if I have a record that only comes halfway the green I turned the input up.

… but you can manipulate it for your own purposes: Hilfreiche Tipps und Tricks rund um mAirList - #6 by UliNobbe

See mairlist.ini, section [Peakmeter].

Standard (in dBfs):

  • Overload=0
  • Headroom=-6
  • BottomEnd=-40

Try your own values, if you like.

1 Like

Thank you Uli for your explanation about the way mAirList handles audio normalization.
It’s useful.

In addition, here are the EBU guidelines for production according to R128

tech3401.pdf (1.7 MB)

1 Like

This is a process we call peak normalisation. Have you heard that back before?

Your example with velocities is quite to the point. But let’s assume, we want to travel with a constant impression of speed. Now we drive our car not referring to the tachometer but our subjective feeling for speed: On the highway we will read 120 km/h, an appropriate speed. On Passo di Stelvio we, appropriately nevertheless, will have a reading of only 45 km/h on the speedo. Same feeling, different levels!

And that’s the essence of loudness metering.

The peak level reading alone is no indicator for loudness. This is because our subjective impression of loudness is, on the one hand, based on the average level, on the other hand on the waveform, i. e. the type of sound, its frequency spectrum (and yet another minor ingredients). Therefore my little example above.

Remember: If you do want to employ loudness normalisation, forget about meters. It’s what you hear.

1 Like

The scale alone doesn’t make an appropriate peak meter. To achieve this, it is inevitable to employ defined dynamics, i. e. the reading with the passage of time. And that is what the mAirList-“meters” are lacking. They are brilliant for indicating qualitatively if there is a signal on the go. Measuring of values should always be left to a dedicated peak programme meter.

1 Like

Like I said earlier I upload a stream and the station don’t ask me to provide the stream in r128 when it arrives at the station it goes through a sound processor.
So I change it back to true peak coming weekend.
is there a affordable software dedicated peak Programme meter or can I use the Orban as one ?

Oh, I missed that point, sorry @henk. Well, in this case you shouldn’t bother about loudness at all, because this only makes sense at the time of producing the show. A complete stream (or audio, or tape …) must not be normalized, because the given loudness differences will not be equalized the whole show only ends up a little lower in level (or, seldom, louder).

Another reason to trash loudness normalisation. Either you go into loudness or you don’t (general remark, I don’t mean you personally). There is absolutely no clue in fiddling with tiny level adjustments and bricking it thereafter.

Hmm… To be honest, I am not up-to-date with software meters as I don’t use them. Users of RME sound devices have the benefit of a free decent software meter. There are some applications around, indded, but I would have to investigate a little deeper. Maybe someone else goes to the fore?

1 Like