Why not use normalize during bouncing?

[ski]

• This is a great thread. Highly informative.

 

• I did a search for "wddq" and could only come up with a radio station in Atlanta.

 

• Next...

Arguing in favor of normalizing the entire mix speaks loudly...

 

Good pun. Still, this is a confusing issue, and disparaging remarks aren't really necessary. Dude!

 

• Makes sense that normalizing audio is a useful function that has its place, but I think the biggest argument for not normalizing a mix is the notion of inter-sample peaks. Speaking of which...

 

Question: do Logic's meters reflect absolute sample levels or do they show "true" level, i.e., inter-sample peak levels? Also, do the meters reflect levels differently at the channel strip level (where they're handling 32 bit audio) and outputs (where they're showing 24 bits)?

[xlm]

I also want to thank everyone for all the great information in this thread. This was exactly what I was hoping for! Many thanks.

[Holger Lagerfeldt]

Question: do Logic's meters reflect absolute sample levels or do they show "true" level, i.e., inter-sample peak levels?

They do not show inter-sample peaks. Stillwell Audio has a 64 bitscope and intersample peak meter for free here:

 

http://www.stillwellaudio.com/?page_id=33

 

http://downloads.stillwellaudio.com/screenshots/bitter_screenshot.png

 

Also, do the meters reflect levels differently at the channel strip level (where they're handling 32 bit audio) and outputs (where they're showing 24 bits)?

I thought Logic actually output 32 bit float to the sound card via the drivers? Though it's 24 bits when you mixdown.

Edited January 18, 2009 by lagerfeldt

[Holger Lagerfeldt]

• I did a search for "wddq" and could only come up with a radio station in Atlanta.

 

I now realize wha wddq means.

 

It's the letters you get when you smash your three middle fingers on your left hand into the computer keyboard.

 

Kinda like going "blah!"

[ski]

Thanks for the info lagerfeldt! I just DL'd Bitter.

 

I thought Logic actually output 32 bit float to the sound card via the drivers?

 

That's a point I'd like to understand better. Assuming that our soundcard/interface is 24-bit... My impression has been that channel strips/instruments/auxes/busses handle audio as 32-bit float, though we never hear it in that format because sound can only be routed to the soundcard via an output object, and the actual output of the output object itself is 24-bit.

 

Is that even remotely correct?

 

Or is it a case that 32-bit float audio info is sent to the soundcard/interface, and the 32-bit float --> 24-bit conversion takes place there?

 

Something tells me it's the former, not the latter, but clarification would be helpful.

[shivermetimbers]

• This is a great thread. Highly informative.

 

• I did a search for "wddq" and could only come up with a radio station in Atlanta.

 

• Next...

Arguing in favor of normalizing the entire mix speaks loudly...

 

Good pun. Still, this is a confusing issue, and disparaging remarks aren't really necessary. Dude!

 

• Makes sense that normalizing audio is a useful function that has its place, but I think the biggest argument for not normalizing a mix is the notion of inter-sample peaks. Speaking of which...

 

Question: do Logic's meters reflect absolute sample levels or do they show "true" level, i.e., inter-sample peak levels? Also, do the meters reflect levels differently at the channel strip level (where they're handling 32 bit audio) and outputs (where they're showing 24 bits)?

 

I didn't mean to belittle anyone and now that i re-read it, that is a neat pun.

 

Discussing the 'N' word (Normalization) may help someone to understand what it is supposed to do, but doesn't mean that someone will know when to use it.

 

Open the hymnals to pg 499.

 

"...Logic Pro finds the point with the highest volume in the current selection area, and

determines how far this is from the maximum possible level. The level of the selected

area is then raised by this amount. The dynamic relationships of sample levels within

the audio passage remain unaltered. "

 

So if I play a Bass guitar and one plucked note is way louder than the general level of the part, how is the passage affected in terms of normalization?

 

Does it bring that Loud 'Pluck' to a normalized level (0dB) and the dynamic range follows but at the same relationship?

Now I play my piano part at pretty much a constant level in terms of volume and dynamics, and normalize it.

 

Comparing the two tracks, my Bass would have the loudest part (pluck) normalized to the 0 level, but that doesn't mean the entire track was set to a 0 level, only the loudest part. The Piano part would have a 'louder' sound because the over all level was pretty much the same and everything is much closer to a 0 level than the Bass part.

 

We have some control over this process in the sample edit, but not in the output bounce process.

 

There is more to mixing than trying to slap on a Compressor or an EQ because we read it somewhere in a forum and hope everything will sound great. Having the good gear will help, but one has to know when and how to use it.

 

 

http://members.tripod.com/No18/tuna.jpg

 

Yes there is a learning curve for some. We see the questions about how checking the 'Normalize' box ruined the mix.

 

DUH! Why did they click that box to begin with? You won't even need it if the mix was done right.

[Holger Lagerfeldt]

We have some control over this process in the sample edit, but not in the output bounce process.

Yes and no.

 

You get worse quality by normalizing in the sample editor as it has to end up in the destination bit format, i.e. 16 bit or 24 bits, while normalization during bouncing is performed at 32 bit float.

[fader8]

Or is it a case that 32-bit float audio info is sent to the soundcard/interface, and the 32-bit float --> 24-bit conversion takes place there?

Not sure. But in our case it's moot as the internal resolution of RME interfaces and Totalmix is 24 bit.

 

When I bounce to 24 bit in Logic, Dither: None, there is a flat TPDF dither present in the bounce file. If I analyze the spectrum from tapping into RME's Totalmix, with Logic streaming live to it, I see the same dither spectrum present, but it's 6dB higher. This is with all gains unity. This doesn't really prove anything other than both mixers require some internal dithering to do their jobs.

 

My guess is the floating point to fixed point conversion happens in Core Audio, and it would seem to make sense the truncations would occurs there too, but I don't know for sure.

 

How's that for an elaborate non-answer!

[fader8]

Here is an interesting footnote to this discussion. I was doing a little testing to figure out how Max DUY worked, and came across this.

 

Apparently Logic's peak indicator fields on its meters have two different states for "0.0"dBFS. Changing the gain 0.1dB toggles between these two states.

 

These plots are with a steady-state 2kHz sine wave and show the state of Logic's peak meter on the output channel. The left plot is 0dBFS un-highlighted and shows no indication of digital clipping or intersample peak indication. The right plot is the highlighted 0dBFS with 0.1dB of gain added and demonstrates the huge addition of clip distortion.

 

Kudos to the Logic dev's for giving us a nice visual warning of this. Otherwise some may think that 0 still means no distortion.

[shivermetimbers]

Logic only works with the 16 and 24 bit info. Right? If my signal goes in at 24 and comes out as 24, what it does internally and beyond my control doesn't concern me.

 

You don't want to normalize to 100% anyway. Get everything recorded right to begin with and there will be very little to monkeyf*%k with later on.

 

I use the bounce function (without normalization) to get my MP3 backing tracks. I don't use the bounce function otherwise. I go from Logic directly to external recording device(s) at 24/48k since that what is what we are stuck with for now.

 

 

 

 

 

My point is that the user should first understand the concept of 'Normalizing' and when it is used. It isn't a bad thing, but it should be and could be avoided by getting things set up properly to begin with.

[Holger Lagerfeldt]

Here is an interesting footnote to this discussion. I was doing a little testing to figure out how Max DUY worked, and came across this.

 

Would you do me a favour and test some files for me? I found some very interesting results during my experimentation with various types of clipping.

 

Does your email work? I tried to emai you.

[fader8]

Would you do me a favour and test some files for me? I found some very interesting results during my experimentation with various types of clipping.

 

Does your email work? I tried to emai you.

Thanks for the heads up. Yes, just checked that account and got your email. Sure, I'll test some files. I'll set up an FTP space for ya and send you the details.

 

Give me a couple minutes.

[jimmymio]

I think you should just ask your mastering engineer what he prefers. He'll give you instructions. Some may want - 3dBFS, or -6 dBFS, and some may be fine with 0 dBFS. In any case, you're not comprimising the quality of your mix.

 

Good advice about asking ahead of time.

FWIW, in his "must read" book, "Mastering Audio", Bob Katz says absolutely no overs and he recommends a max of -3dBFS.

JP

[fader8]

I'll set up an FTP space for ya and send you the details.

 

Give me a couple minutes.

Done. Check your email for my message.

[inquiry]

You're all missing the point that the Normalize option on the Logic bounce dialog, checked by default, is a feature for non-professional users.

 

Remember, "GarageBand Pro" ?

 

Obviously professional users generally won't use it (and don't bounce to MP3 and add to their iTunes Library from Logic either).

 

In sum, if you don't know enough to turn it off, it's best you leave it on. Which is how software defaults are supposed to work.

[Holger Lagerfeldt]

I'll set up an FTP space for ya and send you the details.

 

Give me a couple minutes.

Done. Check your email for my message.

Thanks mate.

[Holger Lagerfeldt]

You're all missing the point that the Normalize option on the Logic bounce dialog, checked by default, is a feature for non-professional users.

 

Remember, "GarageBand Pro" ?

 

Obviously professional users generally won't use it (and don't bounce to MP3 and add to their iTunes Library from Logic either).

 

In sum, if you don't know enough to turn it off, it's best you leave it on. Which is how software defaults are supposed to work.

 

Yeah I get your point... except that Logic Pro has a "Pro" in the title. So it should be off by default.

[inquiry]

Well I completely understand what you're saying. And I hope your finger has recovered from the strain of unchecking it. Once. : )

[NaiveAmoeba]

God, this is getting pretty heavy as a first thread to post on! I usually normalise my tracks when I bounce them, but only because I assume that if I've limited, compressed etc my track properly, the normalisation will essentially have no effect, so there's no loss. But I guess my eyes have been opened!

 

That intersample peak essay is interesting though! Thanks.

[yamahadrums]

I was always wondering why should I normalize when I put so many plugins on a track like compressor, gain etc. It's pointless. Maybe this function is for quick nonproffesional adjusting the level just to have any idea how it will sound in certain circumstances. Otherwise I never use this option when I bounce. There should be some kind of presets for normalising with additional controls. Just thought.

[jbk]

.......0dBFS with 0.1dB of gain added and demonstrates the huge addition of clip distortion.

 

Kudos to the Logic dev's for giving us a nice visual warning of this. Otherwise some may think that 0 still means no distortion.

 

I also checked Logic's meters against the SSL meters and a few others. If your in red '0' your over. They work , they're clear.

[David Nahmani]

If your in red '0' your over. They work , they're clear.

 

Not so. I know people LOVE clear cut rules: if 1 then A, if 2 then B. But if your 0 dB FS digital signal peak is actually at the same value as the peak of the analog signal it represents, then no inter-sample value can be above 0 dB FS, thus there is 0 distortion.

 

Unfortunately it's rarely like that. The only facts you can state with digital audio is that:

 

If your mix peaks between -6 dB FS and 0 dB FS, you may or may not clip the listener's Digital to Audio converter.

If your mix peaks above 0 dB FS, you're clipping.

If your mix peaks below -6 dB FS, you're not clipping.

 

Why everybody is so concerned with that is beyond me, since anyway your final master will peak at 0 dB FS and most probably clip the listener's D to A.

[Holger Lagerfeldt]

Why everybody is so concerned with that is beyond me, since anyway your final master will peak at 0 dB FS and most probably clip the listener's D to A.

Let me tell you why I am - although I doubt my explanation is why "everybody" is concerned about it. If you have multiple stages of clipping during mastering, i.e. clipping digitally just before outputting from the D/A when going to the analog chain, but the level out is at a lower level than full scale, i.e. -6 dBFS, then you can avoid some of the problems. Now, you still get some overshoots but they're not causing headroom exceeding 0 dBFS+ distortion. It also has the advantage of causing less pumping in the analog dynamic processing. Then you come back again thru the A/D and maybe do another bout of (more discrete) clipping, though I prefer not to do that in the A/D. Done correctly it can often sound a lot better and more transparent than normal brickwall limiting, and in a bit of a paradox perceived distortion artifacts are lower. So you still get some 0 dBFS+ problems with the final listener but just 1 time instead of 2.

[inquiry]

God, this is getting pretty heavy as a first thread to post on! I usually normalise my tracks when I bounce them, but only because I assume that if I've limited, compressed etc my track properly, the normalisation will essentially have no effect, so there's no loss. But I guess my eyes have been opened!

 

That intersample peak essay is interesting though! Thanks.

I'm afraid this thread is unduly confusing you.

 

All the normalize function does is find the highest peak in your audio, and determine how far below or above 0dB Full Scale this is. (In the 32-bit floating point math inside Logic, signals can be above 0dB without clipping.)

 

It says "ah, the peak is 3.2dB below 0". So it pushes a (virtual) level slider up 3.2dB.

 

Or it says "ah, the peak is 2.4dB above 0". So it pulls a (virtual) level slider down by 2.4dB.

 

It is identical in all ways, to you listening through the audio, noting on the meter the highest point reached, and manually adjusting the Out 1-2 level slider before you bounce.

 

And if you will then be bringing this bounce back into Logic — all the issues with inter-sample clipping are irrelevant as the level of that normalized track will be getting pulled down again as part of your ongoing mixing and mastering process.

 

The sole sound quality issue is the extremely subtle mathematical effect of repeatedly boosting and cutting the level of digital audio. But again, normalize is doing absolutely nothing that a manual level adjustment doesn't do. If you bounce at 24-bits the degradation will be utterly undetectable to the human ear unless you're doing repeated bounces, with radical level changes of like 90dB.

 

Whether to normalize a track bounce from Logic or not is entirely a matter of personal style and workflow. If you have good control over your levels, you generally won't use it.

[shivermetimbers]

amen

[David Nahmani]

Why everybody is so concerned with that is beyond me, since anyway your final master will peak at 0 dB FS and most probably clip the listener's D to A.

Let me tell you why I am (...)

 

Thanks Lagerfeldt, your explanation makes total sense as usual. As long as you DO realize that we live in an imperfect world anyway, where we can't really control all distortion throughout the entire chain AND stay competitive in matter of levels.

[Holger Lagerfeldt]

 

Thanks Lagerfeldt, your explanation makes total sense as usual.

Thanks, I try my best

 

As long as you DO realize that we live in an imperfect world anyway, where we can't really control all distortion throughout the entire chain AND stay competitive in matter of levels.

Right, that's just not possible. Which is of course why we in theory shouldn't resort to clipping but you know how it works in the real world...

 

Anyway, I just try to minimize the damage, and this seems to be an effective way of reducing the artifacts somewhat while staying ahead of the volume death race.

[CrassMonkey]

About the only time that I would use normalize on a bounce is if the project is mid production and I needed a quick demo of the material. You should EQ everything within the project so that the levels are where you want them before you bounce the project. Normalization will level things out, but not necessarily with emphasis where you want and the results are not always pleasant on the ears. Part of good production is artistic emphasis, and bringing sounds to the front that may otherwise get drowned out by other elements in the song. Normalize isn't forgiving, it just cuts where it see's fit to even things out, and you can lose elements that should get emphasis.

[David Nahmani]

Normalize isn't forgiving, it just cuts where it see's fit to even things out, and you can lose elements that should get emphasis.

No, normalize simply adjusts the level of your mix so its highest peak is exactly 0 dB FS. It doesn't change the mix at all. As "inquiry" said before me, it's the equivalent to sitting there watching your peak meter while your mix is playing and adjusting your master fader so the highest peak is exactly 0 dBFS. That's all it does.

[CrassMonkey]

but, when it is just randomly cutting the DB it is also flattening the mix. To the point on emphasis.. if you are not going through and EQ'ing the items in your mix and relying on normalize to adjust leveling, you are never going to get (or very much not likely to get) the end result you would like.