Batch Auto Gain Bug?

[jmg8]

I just tried using MLoudnessAnalyzer to batch process some kick drums to all the same loudness -16LUFS.
However it turned ALL of my samples up by about 187db!!!

Is this happening for anyone else?

[MeldaProduction]

Hmm, it's actually possible I suppose, if the kicks are short enough. Basically the integrated loudness takes time to stabilize, usually like 10 seconds or even more. Before that it is sort of random, or more like very low value, which would explain why it goes up so much. What we could do is to process the short files several times, so that it all gets stable before processing. I'll see what I can do.

[jmg8]

You are correct, they were quite short files. Thanks.

[pumafred]

In that case, wouldn´t it be almost the same to normalize their peaks? Why concern yourself with integrated loudness of a transient?

[jmg8]

Good point. I didn't think of that to be honest.
I want them to all sound like they are the same level so I just reached for the loudness tool first.
They are much longer than transients though, they are the length of normal kick drum samples.

[MeldaProduction]

Actually normalization is pretty much a pointless concept. The multipass loudness measurement should work pretty well (just tested ).

[pumafred]

Well, then we will be able to normalize integrated loudness

[pumafred]

Still, in something like a kick, I find it strange to use integrated loudness, the longest measurement of all. But, it will be interesting to test....

[mibby]

Actually normalization is pretty much a pointless concept.

I'm curious why you say this? Wouldn't normalizing be a good starting point to make sure nothing is clipping? You could use whatever the gain value is to get to a normalized level as a sort of "maximum gain" level, and not allow anything above that. It's likely I am misunderstanding something here.

[MeldaProduction]

pumafred: Sure, update is on the way . When you think about it, it's just like creating a beat from just the bass drum instead of a single one, so it is actually perfectly fine .

mibby: Normalization had been used for a long time before something smarter was invented, and that's pretty much everything else . From RMS to loudness. Normalization is only useful to ensure nothing is clipping, but that's pointless with floating point 32-bit precision anyways. And when it comes to loudness, there normalization has absolutely no meaning. Imagine 1Hz sine wave at 0dBFS - it has 0dBFS, so it should be loud right? And yet it won't even be audible. So, for many years the normalization is pretty much completely obsolete and useless.

[mibby]

...Normalization is only useful to ensure nothing is clipping, but that's pointless with floating point 32-bit precision anyways. ...

I'm not sure what you mean by this statement. Do you mean that "we should NEVER come close to digital clipping because there is no need to ever record at that level"?

[MeldaProduction]

Hmmm, now I don't understand at all. All I'm saying that in digital domain with floating point precision the level is irrelevant. It may be relevant for processing though, so one should stay in say -20 to 0 dBFS, but if something peaks over, nothing relevant happens.

[pumafred]

Unless you are using some of the analogue modelled plugins, right? They will expect a certain level (-18dFS?), and certainly not past peaking.

[MeldaProduction]

Yes, but still they should work - they need to handle the overs somehow and personally if that means clipping, then it's technically not "analog simulation" anymore, since there's no such thing as overs in analog domain. So it may actually not be necessarily bad .

[mibby]

All I'm saying that in digital domain with floating point precision the level is irrelevant. It may be relevant for processing though, so one should stay in say -20 to 0 dBFS, but if something peaks over, nothing relevant happens.

I am not trying to be argumentative, but I really don't understand. You said "if something peaks over" (I'm assuming 0dbfs) then you have digital clipping. Yes? Why is that irrelevant?