-3dB headroom for premasters - silly ?

How much headroom for pre-masters ?
 
I’ve recently ran into a little argumentation about how much headroom you should leave on your premaster.
 
I want to start with a different scenario:
 
I have a 10Mpix camera.. I want to have the pics printed on paper..
The print shop tells me to shot the pics with a 7Mpix-setting, it will take care of the other 3Mpix in the printing process.
 or:
you want to have a portrait of someone.. everything is set up and you just take the picture of that person plus the other family members that are standing around him.
Because, you have such a high-resolution camera you can digitally zoom into the face and cut it out before you send it out to have it printed.
Now would you do this ? yes, when you think that this person looks especially cute on a picture and you didn’t plan making a portrait.
But if you’d planed to make such a close shot, wouldn’t you have used the optical zoom instead of zooming digitally in ?
(you must be aware of the difference between digital and optical zoom though)
 
 
Now lets come back to topic.. premasters..
You could use the full bitrate/dynamic-scale for your premasters but for some unknown reason you reduce the volume of your projects so that it peaks max at  –3dB, -6dB up to –12dB and even less.
 

When someone recommends leaving no headroom, meaning peaking out to 0dB (without clipping) then there’s a common sense saying that his is absolute nonsense/wrong. In this case it was even recommended to have another mastering engineer do the mastering. (it was his recommendation in the 1^st place to let the client make his project peak at 0dB!)
And you’d assume that this mastering engineer doesn’t know how to do his job (well, it’s ok to be a bit sceptical)
 
Now why would you want to peak at 0dB ?
You still have the comparison in mind that I started this thread with ?
Because you actually have more information (bits) when using the full dynamic scale that is available instead of leaving i.e. 12dB of headroom and losing bits !
 
Now someone would say if you peak at 0dB you’ll not have any headroom.
Well, what is “headroom” anyway ?
Of course you have no headroom if you want to bring up the volume by 5dB with the volume fader.. but a mastering engineer wouldn’t use such a technique anyway. He’d use compressors and limiters and there’s no limitation to how big the signal can be which is  running through those tools.. or don’t you compress/limit your signal when you feed them with signals hotter than –3dB ? Of course you do, and especially a limiter has only been invented to do such tasks.
 
If you have a track with a low level,.. even though I would not recommend it, people just say “hey, normalize that track to 0dB” and no one has a problem with it..
So why would it be a problem having 0dB peak on the 2-buss, and not a problem on a single track ?
 
Back to “headroom”
If my 2-buss peaks at 0dB then it tells me nothing about the loudness of that song... it’s just one peak.. nothing more.. it could be that the next lower peak is at –24db or at –0.2dB.. no one can tell..
So as a general rule, would you recommend to reduce the volume by at least 3dB just to have enough “headroom” ? Don’t think so.
 
The mastering engineer wants to bring up the volume/rms (besides several other things)
Why would it “limit” his work if a signal peaks at 0dB ?
I personally have no problem to compress a 0dBFS signal with a LA2A and increase the make-up-gain by the amount of gain-reduction.. which leaves me again at 0dB (or 0.2dB below it).
This is more or less what a mastering engineer is doing as well, or do you think he compresses the signal and leaves it at –3dBFS ? No, he uses the make-up-gain and brings the mix up to 0dBFS.
 
It’s arguable if it really matters, in a 24-bit environment, if you really hear the difference of a master that came from a 0dB or –6dB premaster… but it’s still true that using the full dynamic range in your premaster that this gives your more bits !
 
I am sure not a “know it all”, and I want to learn these things..
So instead of saying “that all nonsense, you have to have at least 3dB headroom..” I’d like to know “WHY”.
 
Here’s another thread I found about this problem:
[link=http://www.gearslutz.com/board/mastering-forum/36413-how-much-head-room-mastering.html]http://www.gearslutz.com/...ad-room-mastering.html[/link]
 
It’s not the 0dB peaking the signal, the problem is the amount of compression/limiting you’ve done to your premaster. If you have used too much compression, then you might have drastically reduced the dynamics to a point where the mastering engineer has no possibilities to “uncompress” the sound.. and maybe this is the “headroom” some people are talking about.
So yes, leave some “headroom” (dynamics) for the mastering engineer, but don’t throw away those bits !
 
Now you can prove me wrong.

yeah, if it sounds good its good..
I am pretty serious about this.. either I need to convince some people or they need to convince me that premasters can actually peak at 0dB -  this will actually give you a better quality if my theory is correct.

Personally, I'd always leave a little headroom - I usually end up about -6 for peaks.  At 24 bits you ain't really going to lose any info and it leaves a little safety for the mastering process (even if you do it at home).  And it is easy, once the sound part of the mastering process is done to add a few dBs either w/ gain or a limiter.  In your photo analogy, it isn't so much losing information as cropping the image.  Resolution is the basically the same, the ME focuses whats there and cuts down the canvas.

Most ME use hard and software.  The best guy I know uses a TC Finalizer, but he is a wiz and has the grammies on his wall to prove it.  But more use a combo of hard and software - multi-band compression (soft) and hard ware compression/limiting (or maybe software limiting), both kinds of EQ, but usually hardware in the chain.  That takes advantage of the strengths of both formats.

But if the ME asks for a -12 dB mix or -6 you ought to try to oblige him doing the best job.

@

Late to the party, but...

My initial, simplistic take would be that if after all your mixing and effects processing your Master bus happens to be running -6dB with no input or output gain adjustments on the bus, great! Pass it onto the mastering engineer just like that.
But if it's peaking closer to 0dB. Why squash it yourself? The mastering engineer can do that just as well as you. If you have several tracks with gains or trims above 0dB, it might make sense to bring everything down since that's an "artificial" gain, but if all track gains are at or below 0dB, I don't see how you gain anything (pun intended) by "artificially" squashing the pre-master.
On the other hand... I would agree with others that are saying it makes no difference one way or the other with 32-bit files. You'll never hear the difference. So I would just give the mastering engineer whatever he/she asks for, even if it means throwing away a little "natural" amplitude. You'll be squashing the analog noise floor along with everything else so it won't affect S/N. And unless my logic is wrong (quite possible), bringing a 0dB 32-bit file down to -6db would be like having encoded it at 31 bits in the first place. I don't know about you, but I'm pretty sure I can't hear the difference between 31-bit audio and 32-bit audio... especially after it goes through my 24-bit D/A converters that only have the equivalent of 18 bits of analogue S/N ratio, anyway. 

Bottom line, you'll probably end up with a better-sounding project if you don't piss off the mastering engineer by starting a debate about levels. 

I really try to understand this one for all,.. the other thread implied that the ME had no idea of what he was doing when he asked his client to increase the volume by 6dB so that the highest peak would hit 0dB.. and it seems I am the only one thinking the same way as that "poor" ME.

I didn't pay a lot of attention to that thread, partly because I think the 32-bit world makes all that stuff pretty moot. As far as I'm concerned the only place gain staging really matters is in the analog stages preceding the A/D input, and in not clipping your A/D converters on recording. Other than that, and maybe keeping plug-ins from running in the red where some may not behave well, I think you just want to avoid radical gain manipulations and pegged meters that don't give you any information. 


 By "squashing", I just meant if you had a project that was running close to 0dB on the Master "naturally" (i.e. without input gain on the master or make-up gain on a compressor or a lot of output gain on the buses feeding the master, etc.), and you pull the bus output gain down so the exported file is at -6dB per the mastering engineer's requirement. That would be an "artificial" gain reduction that the engineer could do just as well, starting with a 0dB file, and setting the track trim in the mastering environment at -6dB.


But again, I really think it's a non-issue with 32-bit files. In 32-bit world, -6dB is nothing in terms of lost resolution.

what i think you're saying, is that you let the master peak all the way to just under 0db....

without any buss compression or whatever, just your natural mix, is coming close to 0db on the main buss out, correct?

that makes sense to me.

i see that in my own mixes all the time, where i'm not using ANY compression anywhere, except maybe what i applied during tracking, and i get the odd snare or kick or collective that will drive the one peak WAY UP.....
while the rest of the time, the bulk of the mix sits comfy around -16 to -12 on the master meters.

Did you ever ask the ME why he recommended normalizing to 0db peaks? I find that surprising. Most MEs would prefer that you give them a lot of headroom to work with. Nevertheless, peaking at 0db isn't necessarily a problem. Your average RMS is much more important to the ME.

Imagine a transient spike, such as a snare hit that rises to -3dbFS. Is that going to stand out and get your attention? More or less so than a snare hit that hits 0dbFS? Well, it depends.

It depends not on the absolute level it hits, but how high it goes relative to what came immediately before it and what comes immediately after it. If the average level of the surrounding music was at -14db, then that -3db peak represents an 11db rise and is very much present and in your face. But if the average level of the surrounding music was, say, -6db, then even the 0db-peaking hit will be wimpy by comparison, even though it went to a higher number.

So if you have a very dynamic mix, allowing occasional peaks to hit near 0db won't necessarily be a problem. But it could be, and I wouldn't be surprised if the ME actually turned down the input levels to his mastering limiter specifically to restore some headroom so his compressor had something to work with.

I'd also like to submit that the oft-repeated idea that maximizing levels "uses more of the bits" is bogus. Yes, you do use more of the high-order bits, but that means nothing. There is no audible advantage to turning on the high-order bits.

Yes, you will raise the SNR (at least with relation to quantization noise), which sounds like a good thing. But remember that 24 bits gives you a ~144db dynamic range - but human ears only have a (theoretical) 120db range, the practical hearing range is only about 80db, and a typical well-mastered pop/rock tune has about a 14db range.

The benefit of exercising 2 or 3 more bits is like scraping the paint off your ceiling so it'll be further from the floor.

A couple of points here. Firstly my favourite old subject and that is K system metering. Now if you produce and mix a track based on the K -20db standard then you will end up with a mix that has an overall RMS value around -20db but with peaks that may reach -3db at best.

If you produce and mix around the K -14db standard then you will end up with a higher RMS value (6db in fact) of around -14 db with peaks just reaching -3db or so. What I am saying here is if you standardise your metering you will never go wrong.

I am bit wary of the term mastering engineer. I am sorry to say that there are people who call themselves this but do not necessarily do a great job of mastering and they demonstrate that by saying things like just give me the mix up to 0db. That is just so wrong. I have mastered hot mixes myself and it is a pain and yes you have to end up lowering the input level to the first process in order to get things back to normal.

So the advice here is dont push things too hard in the mix. There is no harm in applying a bit of premix compression just to hear what happens as a result of compression later. This is OK to do. Because sometimes you have to alter the mix a bit to accomodate the later mastering. You can switch this premastering compression off just prior to printing your mix.
I sometimes leave it on too but the compressor has a high threshold so it is only gain reducing 3 db max and the ratio is low eg 1.3 to 1. The attack needs to be a little slower to to not kill any transients in the mix.

I had a bad experience recently. I did a mix of a hip hop track for a compilation CD but was not allowed to master it and that was fine by me. But when the CD came out the ME did an absolutely terrible job of my great mix. He just slammed it and overdid everything and the song actually got softer in the loud bits. He totally ruined it. I could have mastered it a million times better and got it loud and crankin without any artifacts so the moral is not all ME's are good. Band was very unhappy too. I gave them a demo of what I could do and they were stoked.

If you are going to send stuff out to a ME, get some examples of their work first and give it all a serious listen and then make up your mind.

Just slightly off topic...

Today I was driving in the car when Green Day's "American Idiot" came on the radio. Whoa! It sounded just AWFUL. I've heard better on the Songs forum from guys who introduce their tunes with "this is my first attempt...".

That same song on a CD is much, much better sounding. Still squashed to within an inch of its life, but appropriate for the genre and not unpleasant to listen to (although I've never been able to play the whole album all the way through in one sitting).

The problem is that when you smash it that hard on the CD, the radio station is just going to smash it again and turn it into unrecognizable roadkill.

The next song they played on that station was a stark contrast - it was a track from Dark Side of the Moon. Still annoyingly squished by the broadcaster, but the clarity was sooo much better. I wonder if producers who insist on maximum loudness from their ME ever listen to their own stuff on the radio.

the ears decide.

But, for me, to be safe, I like -3 to -6 db space, so I am sure that nothing peaked at all. And, it is a fast way to get it to a level where it is ready to master, instead of trying to get it perfectly where it peaks exactly at 0db.

When they say -3 to -6db, they dont mean the highest peak is exactly in tat place. They just mean in that general area, because it is way less likely to have any peaks in it that they cant deal with.

I see your point, I don't think it really matters if the analogy refers to color, detail or size of a picture. You wouldn't want to have a degradation in neither of those points.

The number of bits will always be the same, if you turn the master fader up or down..
it's the amount of "ones" that change.
Now keep in mind that if you bring your project up to 0dB (without the use of compression) just by using the volume-fader(s), you don't squash your dynamics, you'll get more of the "ones".. more "bits".. more valuable information (instead of reducing the volume and getting more "zeros")
The problem is indeed the "valuable information". Does more of the "1" - bits mean more "good" bits ? not necessarily ! Maybe it's even getting to the worse, but this amount of degradation is inaudible.
Do I still have enough headroom when you peak at 0dB ? YES ! (asuming you didn't totally compress your mix, which is a separate point.)
Now would it matter if I or the ME increases/decreases the volume of the project ? hmm...
If I change the volume in my original project, then the sequencersoftware can re-calculate the new volumes through it's internal processing, those signal-"routings" are not available after mixing down to a stereo-file.. so, it could make (a little) difference.
I haven't come across any other thread yet dealing with that 0dB / -3dB / -6dB premaster problem.. I know there are plenty..

I think it's like this: If the mastering engineer felt like he/she wanted to start with some EQ boost pre comp he/she would have to first lower the gain before applying EQ. Also as mentioned if he/she wanted to start with some expanding or parallel compression it would clip if gain wasn't lowered pre to applying processing. Also without the headroom the ME would have to go back after each succesive stage and check that nothing has clipped and if something has then all stages after that stage would have to be redone if something was missed in the first place. Since there is realy no need for having the mix hot (maybe unless your at 16/44.1) it's just unnecessary evil

Also having the mix peaking at 0 or just below might leave it with inter sample clips in the digital domain if not fp and also it's not sure that the ME will work in digital so those inter sample overs would/could cause minor problems with saturating the analogue channels.