Debunking myths - DAW Mix Bus Levels

[Ethan Winer]

Several times recently I have seen the advice to keep DAW tracks low in level, rather than let their sum exceed zero which requires reducing the master bus output to compensate. This is so easy to test that I just did it, and the results are exactly as I expected. It makes no difference at all. Here's my report:

I created three 24-bit wave files at a level of -4 dB in Sound Forge. One each at 100 Hz, 1000 Hz, and 4500 Hz. I loaded them into Sonar on separate tracks and set each track's volume to 0 dB. When I hit Play the bus output clipped, registering +5.4 dB. So I lowered the master bus volume to -10 dB, which prevented clipping, and exported the mix.

Then I set each track to -10 dB and set the bus output to 0 and confirmed the same playback level. I exported that mix and used the FC utility to compare both files. They were identical.

So this proves the master fader in Sonar properly compensates for tracks that exceed 0 dB when summed. It also disproves claims that you need to be as careful with "gain-staging" in a DAW as when mixing through an analog console. At least with Sonar anyway.

--Ethan

[max cooper]

Originally Posted by Ethan Winer

So this proves the master fader in Sonar properly compensates for tracks that exceed 0 dB when summed. It also disproves claims that you need to be as careful with "gain-staging" in a DAW as when mixing through an analog console. At least with Sonar anyway.

--Ethan

Nice work.

But there's another issue that's been debated recently regarding not hitting the A/D as hard.

For example, when I'm tracking a vocal and I'm hitting close to zero vs -3.0 or so things sound smoother and more dynamic.

I've had the same result with Rosetta 800, Fireface and the bad ol' digi 002 rack converters.

I'm not saying this jives with conventional wisdom about how these things are designed, but I do hear it. Yes, I should test it to be sure, but I do trust my ears.

I would suppose the analog section of a good A/D would be more than capable of delivering 'nice' audio up to zero in the digital section, and I would doubt the digital section cares about anything other than overs, so I can't really guess where the disconnect is.

[Ethan Winer]

Max,

> But there's another issue that's been debated recently regarding not hitting the A/D as hard. <

Sure, though what I'm talking about happens before the output D/A converter. That is, it's the "summing" math the program applies to determine what to send to the D/A or to write to a mixdown Wave file.

--Ethan

[tomdarude]

and another issue that always comes up in this discussion is,
that some plugins sound worse if they´re hit to hot !
so it´s not exactly about the "pure" mixbuss-headroom, but if the plugin has
"headroom"-issues....especially EQs (maybe they´re not adressed in 32bit???)
so that would be more about gainstaging inside the channelstrips itself, rather than the summing.


does anyone have real scientific information about this??
is there a list of plugins that have such restrictions in theory ? (if that´s true!)

how about the good sounding and/or DSP-based plugins (URS-api/UAD-1073/Duende)
did you experience any gainstaging-issues in any DAW´s with them ??


curious too.....and thx for the test ethan, I always felt comfortable lowering the masterfader so far (in Logic that is) thumbsup

[briefcasemanx]

not only the analog stage of your converters, but the preamp as well. I keep tracks around -10 so that they're somewhat close to analog unity.

[Hexfix93]

well you can drive the channel way beyond what the meter shows, then you gotta use your ears and not the meters...

in the end, i find my mixes work out best when i keep the master at 0. all programs handle headroom differently. you cannot clip fix point stuff or it sounds awful.

I like keeping low track mixes recorded in at no hotter than -6db, and mix everything around -6db, i seem to get the best results because it helps me visualize with the faders and leds better and i know visually where my sound is at and that helps my ears in the end, helps me know exactly what i am doing. once you start pushing levels beyond that it gets harder to tell visually whats going on. at least thats how i see it.

if i have a high track count i record in at no hotter than -10 or - 12 db and mix the master to about -6db, but keep my tracks at about - 10 to -12 on the things i want loud. these rules have helped me immensely in my recent projects, and i am finally getting a sound out of sonar 6 that i actually like a lot. and its impressing me all the time...

If i want bass to be up front, i compress with a good hardware compressor. that works great too as long as i record no hotter than -6db on low track counts and -10 and -12 in high track counts....

[theblue1]

Amusingly, just a couple days ago I performed some quick tests parallel to those that Ethan did (also in Sonar) just because I wanted to confirm to myself that -- at least in a simple one track/no plug situation -- that Sonar's floating point engine treated "localized overs" in the chain as expected.

But, as noted above, not all plugs may handle overs gracefully, even if they, too, use FP math, since they may well be optimized for "normal" levels. If you have a non-over-friendly plug across the bus you might well end up overloading it, seems to me.


And, though I'm not sure how it all works out, it's worth noting that not ALL DAWs even use floating point math. As I recall PT uses 48 bit fixed and I think there may be one or two others. Like I said, I'm not sure of the implications -- but I'd be delighted if someone could illuminate the ish...

[Ethan Winer]

Tom,

> some plugins sound worse if they´re hit to hot ! <

Could be. My simple little test was with no plug-ins. As far as I know the Sonitus plugs that are my mainstay are 32 bits internally. I've had mixes end up all over the place - tracks high, master low, and vice versa - and I've never noticed a problem. Then again, I've been an audio pro for years, so my levels are never totally crazy...

But the main point of my test was to show that yet again "conventional wisdom" is wrong. This was so incredibly easy to test I'm surprised everyone hasn't tried it.

--Ethan

[John Suitcase]

I've had similar experience in Cubase SX, in that turning down the master buss eliminated clipping.

But, just the other day I had a track where I had used a plug-in (the included compressor plug-in) to make-up some gain on a guitar. During mixdown, I could hear it clipping, though the meters showed it being fine, both at the track and the group channel that I had the guitars going to. Eventually, I discovered that boosting the gain inside that plug-in was causing clipping somewhere, so I lowered that make-up gain, and bussed that guitar to a different group, which I could bring up by itself. This solved the clipping problem, and showed that even 32bit has a limitation, probably where a plug-in hands the data back to the application. I don't know the details, but that was my experience....

[mplancke]

So this proves the master fader in Sonar properly compensates for tracks that exceed 0 dB when summed. It also disproves claims that you need to be as careful with "gain-staging" in a DAW as when mixing through an analog console.
--Ethan

It's more complicated than what you've eluded too. I've heard a distinct difference in the sound of my productions when I've lowered track levels. IMO it has nothing anything to do with overdriving the mix bus. I attribute the improved sound quality to not overdriving plugins. Not all plugins handle extended levels in the same manner, especially critical on the master bus.

Paul Frindle made an excellent post on why it makes a difference.

http://gearslutz.com/board/showthrea...t=paul+frindle

YMMV, always use your ears to decide what works best for you.

Mark

[Bob Olhsson]

Gain staging is where all of the problems lie. Many of us have the experience that in 24 bit digital it's far better sounding to error on the low side than the high.

"Just use your ears" is unfortunately a case of analog thinking in the digital domain. Digital clipping can be very hard to hear however its effect is to create erroneous numbers that turn audio into a brittle edgy mess if they get recalculated later on as part of some additional digital processing. Unfortunately one isn't always in a position to fix bad math after the fact.

[Ethan Winer]

Mark and Bob,

> Not all plugins handle extended levels in the same manner <

Thanks, point taken. When Iu get a chance I'll try the same thing with some of the plug-ins I use engaged.

--Ethan

[max cooper]

Originally Posted by Ethan Winer

Mark and Bob,

> Not all plugins handle extended levels in the same manner <

For example, the Digi EQIII that has earned so much praise is a real crapper at the top end.

URS too.

[sleepwalker]

I've performed similar tests using normalizing up then down and have had files null completely.

There are some who say moving a fader in a 32float DAW will degrade your audio. Ethan simply debunks this very simple statement. Is gainstaging important in DAWs? Sure, but it's actually extremely complicated. Automation and plugins open up a whole new realm of complication.

[MWP]

It's probably just me, the digital PT DAW noobie, but here is my my experience.
I like to mix at approx 80-85 db and I try to get my mixes at -18rms. When mixing songs with multiple guitar and vocal tracks, I find that I have to lower the faders on the individual tracks, and then crank the Master fader up, to get what my ears want to hear. Yes, I am using plugins (ex Eq, Reverb and compressors), but I do not understand why just doing the opposite would change the sound so differently?

MWP

[Ethan Winer]

Originally Posted by MWP

I do not understand why just doing the opposite would change the sound so differently?

Not to dilute this thread with more of my own myth-busting, but this article explains some of these kinds of perceptions:

www.ethanwiner.com/believe.html

I'm not trying to suggest what you are or are not hearing! Just pointing out that what we think we hear is not always accurate.

--Ethan

[starseed]

Originally Posted by Ethan Winer

Not to dilute this thread with more of my own myth-busting, but this article explains some of these kinds of perceptions:

www.ethanwiner.com/believe.html

I'm not trying to suggest what you are or are not hearing! Just pointing out that what we think we hear is not always accurate.

--Ethan

Good one, thanks.

I have been using some panel traps made to your measurements (in combination with a purpose built room) and I am most pleased with the results. 0.3 sec over the whole spectrum. Opened my ears right up.

I agree, only after the needed measures have been taken does a cable make any degree of difference.

[Paul Frindle]

Originally Posted by Ethan Winer

Several times recently I have seen the advice to keep DAW tracks low in level, rather than let their sum exceed zero which requires reducing the master bus output to compensate. This is so easy to test that I just did it, and the results are exactly as I expected. It makes no difference at all. Here's my report:

I created three 24-bit wave files at a level of -4 dB in Sound Forge. One each at 100 Hz, 1000 Hz, and 4500 Hz. I loaded them into Sonar on separate tracks and set each track's volume to 0 dB. When I hit Play the bus output clipped, registering +5.4 dB. So I lowered the master bus volume to -10 dB, which prevented clipping, and exported the mix.

Then I set each track to -10 dB and set the bus output to 0 and confirmed the same playback level. I exported that mix and used the FC utility to compare both files. They were identical.

So this proves the master fader in Sonar properly compensates for tracks that exceed 0 dB when summed. It also disproves claims that you need to be as careful with "gain-staging" in a DAW as when mixing through an analog console. At least with Sonar anyway.

--Ethan

You are correct - the idea that digital systems cannot add up is one of the most bizarre amongst the many in our industry. I have not found a system yet which clips sample values greater than notional +/-1 in it's summing buss, or one that has such low accuracy that at 24bits a difference would occur with signals -10dB below max.

As I have said in other threads before - I suspect this myth comes from the mixing 'in or out of the box' debacle, which of course is an entirely different thing and nothing to do with adding up..

[Methlab]

Im starting to peak at lower levels then I did before when i mix.

I noticed while playing back a vocal that I tracked inCubase that I heard distortion in one part where the singer got louder. I went back to her track and listened, and even looked for square waves because I didnt hear it solod. I then played the doubled vocal and didnt hear it. I thought I was nuts because I had tracked in with compression and I knew nothing was clipping, and the peaks were controlled.

I then played them both together and heard it. I went and looked at the 2 buss and sure enough it hit the red and sounded very bad. I then adjusted the faders and it all went away.

So, in this instance, I heard it clip inthe 2 bus and it did not sound good.

[Ethan Winer]

Hi Paul,

Thanks for chiming in.

> the idea that digital systems cannot add up is one of the most bizarre amongst the many in our industry. <

There are a lot of very bizarre notions out there, as I'm sure you know. My recent "believe" article above is my small attempt to explain how such myths might develop. Then again, some people will not accept a logical explanation no matter how compelling it is. I guess that's why they call these discussions "religious."

--Ethan

[Paul Frindle]

Originally Posted by Methlab

I then played them both together and heard it. I went and looked at the 2 buss and sure enough it hit the red and sounded very bad. I then adjusted the faders and it all went away.

So, in this instance, I heard it clip inthe 2 bus and it did not sound good.

Yes what you were seeing is combinational overload with a signal that was simply too loud - not a clipping of the mix bus itself..

This subject has been discussed again and again - but the simple fact is that the way the W/S apps are gain scaled (without actual signal headroom wrt to REAL signal) and the way they present themselves, always causes confusion and encourages people to overmodulate.

This is a naivety in the way such apps have been originally designed, mostly from the user interface perspective, which has now become the norm - it's not a technical summing error as such and IMO it's not the users fault either :-)

It's a case of the 'new kids on the block' not heading what the 'old farts' had always understood. It's sad and completely avoidable - as are so many issues we fight with daily, often without even knowing they are there :-(

[MWP]

Originally Posted by Ethan Winer

Not to dilute this thread with more of my own myth-busting, but this article explains some of these kinds of perceptions:

www.ethanwiner.com/believe.html

I'm not trying to suggest what you are or are not hearing! Just pointing out that what we think we hear is not always accurate.

--Ethan

Thanks Ethan, interesting article and no offense taken.

MWP

[Paul Frindle]

Originally Posted by Ethan Winer

Hi Paul,

Thanks for chiming in.

> the idea that digital systems cannot add up is one of the most bizarre amongst the many in our industry. <

There are a lot of very bizarre notions out there, as I'm sure you know. My recent "believe" article above is my small attempt to explain how such myths might develop. Then again, some people will not accept a logical explanation no matter how compelling it is. I guess that's why they call these discussions "religious."

--Ethan

I agree. This is technology and it shouldn't be about beliefs. The point is to hear what happens and find out exactly what is causing it methodically and intelligently. It's true that what we think we hear so often turns out to be due to something apparently illogical and unexpected - or even something very important, but so obvious we can miss it completely it we are not careful.. And this last effect happens most often when we think we understand it already and we have long since stopped questioning it..

[theblue1]

Originally Posted by Ethan Winer

Not to dilute this thread with more of my own myth-busting, but this article explains some of these kinds of perceptions:

www.ethanwiner.com/believe.html

I'm not trying to suggest what you are or are not hearing! Just pointing out that what we think we hear is not always accurate.

--Ethan

I already had a real sense of the realities behind your eye-opening monograph, Ethan, but seeing the measurements really brought it home. I've always used a set of visual and other cues for positioning my ears in an attempt at a standardized sweet spot... but your measurements show what an uphill battle it is.


And I have to say that I had really not considered small locational differences as a factor in listening tests... which seems like a real head-slapper when you look at some of your plots.

I think it was just this sort of endlessly unfolding quest thang that finally allowed me to bust out of being a "hi fi nut" when I was a kid and start listening to music, again.

Happily, I had that earlier obsession to measure against when I was building my project studio rig in the 90s and headed myself off from any unattainable grail quests. Lo fi... it's affordable, it's hip... and it's so attainable...

[Paul Frindle]

Originally Posted by theblue1

I already had a real sense of the realities behind your eye-opening monograph, Ethan, but seeing the measurements really brought it home. I've always used a set of visual and other cues for positioning my ears in an attempt at a standardized sweet spot... but your measurements show what an uphill battle it is.


And I have to say that I had really not considered small locational differences as a factor in listening tests... which seems like a real head-slapper when you look at some of your plots.

Ethan's article is a good illustration of the kind of thing that can happen to convince someone that an immediate difference occurs when something insignificant is changed. And it is true to say that comparative listening tests should be conducted using an A/B switch in a constant listening position.

However the room plots can be misleading since the response is caused by reflections around the room which we naturally become accustomed to. In other words, it does not sound anything like as bad as the plots would suggest in reality. Now this in itself IS THE most interesting part of it, since it bears witness to just how much 'processing' goes on in our brains and to what degree we naturally compensate and take account of environmental cues - that's what evolution has given us.

Now therefore, one massive complication is that anything that disturbs our ability to do these natural corrections interferes with our compensating mechanisms and risks letting us hear some of the room response as it really is!! Now that is another long subject - but you get what I am driving at... There are classes of technical errors that although very small in themselves can none the less create unexpectedly significant sonic effects - well beyond the sensitivity normally associated with the hearing process. And our sensitivity to them is significantly greater when listening in a normally reflective acoustic environment than if listening in a highly treated room or with headphones.. This is one of the many paradoxes that people have difficulty comprehending when they try to equate measurements with what we actually hear.. Fascinating subject, I have spent decades exploring.

And this is why paradoxically - if you are doing professional audio work where artistic decisions are being made, aiming for a perfect measured room response is not actually in your best interests - despite what everyone believes.

[theblue1]

There are classes of technical errors that although very small in themselves can none the less create unexpectedly significant sonic effects - well beyond the sensitivity normally associated with the hearing process.

Paul

I'd be very interested in reading more about these unexpectedly significant sonic effects that are well beyond the sensitivity normally associated with the hearing process...

As someone with a little (not a lot, mind you) background in the metrics of perception I'm immediately, let's say, intrigued when I hear about modes of perception which science hasn't been able to measure in the last several hundred years. Mind you, I think it IS a big, mysterious world -- but I like to keep the Mystery column and the Science column on separate parts of the ol' spreadsheet of knowledge to avoid one sneaking into the other.

If you can point me to any supporting science for that, I'd be fascinated, I'm sure.

[Methlab]

Paul,

What level do you think a mix should be like in a DAW environment? Like where should it be peaking out at ?

[sleepwalker]

You heard it clip the DA converters. TOTALLY different. Drop the Soundcard output in Cubase and all of the sudden your 2bus stops clipping.

Originally Posted by Methlab

Im starting to peak at lower levels then I did before when i mix.

I noticed while playing back a vocal that I tracked inCubase that I heard distortion in one part where the singer got louder. I went back to her track and listened, and even looked for square waves because I didnt hear it solod. I then played the doubled vocal and didnt hear it. I thought I was nuts because I had tracked in with compression and I knew nothing was clipping, and the peaks were controlled.

I then played them both together and heard it. I went and looked at the 2 buss and sure enough it hit the red and sounded very bad. I then adjusted the faders and it all went away.

So, in this instance, I heard it clip inthe 2 bus and it did not sound good.

[Paul Frindle]

Originally Posted by theblue1

Paul

I'd be very interested in reading more about these unexpectedly significant sonic effects that are well beyond the sensitivity normally associated with the hearing process...

As someone with a little (not a lot, mind you) background in the metrics of perception I'm immediately, let's say, intrigued when I hear about modes of perception which science hasn't been able to measure in the last several hundred years. Mind you, I think it IS a big, mysterious world -- but I like to keep the Mystery column and the Science column on separate parts of the ol' spreadsheet of knowledge to avoid one sneaking into the other.

If you can point me to any supporting science for that, I'd be fascinated, I'm sure.

I realise that this kind of thing is not something one can say in public without inviting conflict. I have had more conflicts on this level than I can remember - but that's another story.

It's not that these things cannot be demonstrated, or that attempting measurement and quantification is somehow pointless or contrary to some bizarre doctrine (heaven forbid) - it'd just that quite naturally what we perceive cannot be so accurately measured. And in a technical culture where measurement most often rules (because it is 'doable', quantifiable and justifiable), the more nebulous factors of how we actually perceive are sadly often overlooked (whether or not there is existing 'science' to refer to). This means that sometimes we are measuring the wrong things or looking to the wrong phenomena to explain the perceptions we experience (much as Ethan's essay outlines). Or we may make mistaken technical decisions based on incomplete existing science and understanding etc.. I presented an AES paper on this subject around 10 years ago, but my feelings on the subject are far deeper than the paper suggests - for obvious reasons in a public presentation :-)

The fundamental problem is that once we delve into aspects of how we percieve, subjectivity is firmly on the agenda - and we must not forget that every experience any of us has to any sound, by the time it gets to our conciousness IS necessarily and forcibly subjective.

This rather obvious issue I have quoted with the listening space paradox is but the tip of the iceburg IMVHO and long experience of these things. All of my work is the result of designing applications and sound products for strictly commercial purposes - so none of this work was bounded or limited to existing published science - I was just after the effect and had to understand what provokes us to hear it and how to reproduce (or suppress) it , regardless of whether there was existing science or not, or whether what I experienced conformed to that science or not... I have never been part of a research organisation that published peer reviewed papers, nor do I have the credentials to be credible in such a fraternity. But what I have found out over the decades I use everyday in the products I design. Learning is an ongoing process for me that has spanned my whole career and is not about to stop. I am prepared to share it with anyone who's interested - as always - because it fascinates me too :-)

I should add that I'm absolutely not about debunking myths, for the purposes of denigrating others or denying what they experience. What I am interested in is understanding from where these myths come, why they are important to people, how they perceive the problem, understanding why they perceive it and what causes it, so that I can learn more useful stuff for more useful purposes.. I have learned much from people who reported hearing stuff I would not have given credit for - had I not retained an open mind and due respect for others :-)

Is there any particular issue of interest you would like to raise? I will do my best to give a coherant and honest opinion :-)

[Paul Frindle]

Originally Posted by Methlab

Paul,

What level do you think a mix should be like in a DAW environment? Like where should it be peaking out at ?

Well if you are talking about the final mixed output? The problem is that the meters on DAWs show only sample value and not signal - the reconstructed output signal the destination DAC will need to cope with can be up to typically 3dB higher than this in parts of some programme (or even more under very extreme and contrived conditions).

This suggests that aiming for around -3dB would make it safe under most normal conditions - using the DAWs usual meters.

However if you have a reconstruction meter (i.e. an application that specifically shows you the signal level rather than sample values only) you can mix to the level that satisfies that meter with safety - and therefore gain some modulation for programme that did not cause overs in the first place.

Another approach is to get an application that will dynamically correct only during the periods of the overs - this will ensure conformity without you having to worry about it or lose level throughout the whole mix etc.. However, obviously such an app may change the sound of the programme during the passages that create the overs, so being aware is still a good idea..

Sorry - forgot to say::

If you are talking about levels within the recording and the DAW during the mix and between plugs etc? I would aim to lose around 10dB at the start of the channel and try to keep levels between plugs below around -6dB..