What is normalization?

[PPH]

I did however notice that normalized file sounds a bit... mmm... different. I try to not normalize sounds.

Maybe the normalization algorithm of Modplug is not good (I don't know; I remember people complaining; maybe they were wrong; maybe it was fixed).

You can always use another software for normalizing (for example, Audacity; in Audacity, though, the "Normalize" function does not make the volume as loud as possible; it leaves some room; but can use the "Amplify" function, which lets you normalize the sound).

[Saga Musix]

Yes, people always rumour about "algorithms" that are oh-so-complicated. Why don't you just check it yourself? MPT is open source. It's one simple damn multiplication. No noise shaping, no nothing. Just like normalization is supposed to be. How could you do this differently?! Also, this "-2dB headroom" thing does not make sense with samples. It's only useful for final mixes which are usually rendered at -2dB. But for 8-Bit (not so much for 16-Bit) samples, a -2dB headroom would be fatal.

[Sam_Zen]

I don't consider normalization having an algorithm either. It's just a plain simple math formula.
Compressions maybe have algorithms, but that's not relevant here.

[psishock]

I always thought normalization and maximization are different things. Afaik, normalization is when the volume is evened out. Maximization is when it is maximized - all of what you guys said in this thread. I might be wrong, but GoldWave certainly made such a difference in older versions.

Goldwave is my first wave editing tool, maximization == normalization there, that is for sure, i am using it for several years now. It has a separate compressor tool and i'm using it too for pushing loose high peak levels down in the final mix, but those are two different things. Maybe they had the two tools together in (really) older versions.

BTW: Awesome new avatar, psishock.

Thanks , we were on some open air party and i've noticed a hat on my friend. I definitely wanted to try it out on me, and the last avatar was around 5 years old, so it was a good opportunity to change.

[g]

It's one simple damn multiplication. No noise shaping, no nothing. Just like normalization is supposed to be. How could you do this differently?!

I don't think it's that preposterous to think that normalization algorithm could have dithering. There will always (well almost always) be rounding errors, and when it comes to audio a lot of people will always go for overkill

[Saga Musix]

I think it's very unusual (at least with samples - mastering is a completely different story) - correct me if i'm wrong, but i've never seen a program that combines normalization with noise shaping.

[g]

I have no idea I agree it is rather pointless (when not mastering), I just wanted to make a point that it's not a silly idea to think that a normalization could be a fairly complex algorithm.

I do however think a gain factor which didn't actually affect the sample other than at playback. I guess it would be like the old pre-amp value except sample based. And it would break all compatibility as we know it. Still, non-destructional editing FTW! Perhaps in OpenMPT 4.

[machinesmith]

*runs back again*

Suprisingly my source check for info actually came from Goldwave's Help file (and PPH's explanation was near verbatim)!

I'm just curious though, Louigi I checked Goldwave ( v4.26 - I'm on win98 so Old is necessary) and if you go to `Effects/Volume/Maximize' the tooltip displays: `Maximizes (normalizes) volume' which version are you using ?

Also:
Harbinger did your question get answered?

*runs... o never mind*

[Louigi Verona]

I thought it was version 4.26... perhaps it is my wrong and I have mixed things. Anyway, I did think in OMPT that when you export a song with normalization it sounded differently - a bit more flat on the dynamics.

[PPH]

Yes, people always rumour about "algorithms" that are oh-so-complicated. Why don't you just check it yourself? MPT is open source. It's one simple damn multiplication. No noise shaping, no nothing. Just like normalization is supposed to be. How could you do this differently?! Also, this "-2dB headroom" thing does not make sense with samples. It's only useful for final mixes which are usually rendered at -2dB. But for 8-Bit (not so much for 16-Bit) samples, a -2dB headroom would be fatal.

I was afraid someone would take offense. That's why I made it clear that that was not my opinion. I was merely looking for an answer to Louigi's comment.

I agree with you. I know normalizing is a simple multiplication. And I'm not saying that I would have it done like Audacity does (I never use Audacity's normalize function).

EDIT: and I did not check myself, precisely, because I didn't notice any problems.

[Saga Musix]

no offense taken, i'm just.... allergic towards stupid ideas.

[Harbinger]

Very thorough explanation. Especially by LPChip and pictures by Sam_Zen. All of you did well to clear it up. I guess "normalize" means "amplify every sample until the sample with the highest peak reaches the top". But "amplify" means "increase the amplitude of every sample by a set amount, no matter what, limited only by the highest setting."

Now my next question is: When would you NOT want to normalize? And if never, why isn't it done automatically to imported samples?

[Saga Musix]

You would want it in some formats which don't support global volume for example, or if you want to use your samples in another way than directly playing them in the pattern - like, if you use mod music in your game and use the mod samples for ingame sample playback - no this is not weird, this is how it's done in many old games. So this is one of the reason why you can't force anyone to use normalization.

[Sam_Zen]

I don't like being forced to automatic functions anyway. The user should have the control and the choice.

When would you NOT want to normalize?

It could be that you have prepared a set of samples, each with its own defined volume in relation to the others.
Normalization would disturb this setting, and you will have to add extra volume codes with each note to correct the relations again.
With several tracks for an album it's the same. Sometimes one wants 1 or more tracks being softer than the rest.

Another reason to NOT normalize : the moment in the process of sample editing.
I stick to the rule : normalization always as last modification.
If you start to normalize, and then apply e.g. some filter to boost some frequency range, big chance of clipping.
The same is valid for other effects as well. Normalization should be seen as a last enhancement of the result.

Tip : if adjustable, never set the normalization factor at 100 %, just leave it at 96 or 98 %.
The hearing-difference can't be noticed, and it's just to be safe.
To be sure that, during some conversion, mix or such, some peaks get a slight higher value due to calculations.

[Harbinger]

Excellent excellent advice. These tidbits will find their way into the Help Manual!