<insert yawn emoticon>

Berlin Dubplates and Mastering (one of the most frequently appearing names in the credits on my records) have an FAQ on what to provide here: http://www.dubplates-mastering.com/faq.html

No mention of peak levels.

Rashad Becker of BDaM (probably one of the most in demand mastering engineers in the world if my record collection is anything to go by) was interviewed by Monolake here: http://www.monolake.de/interviews/mastering.html

Very enlightening reading!
Last edited by cron on Sun May 13, 2012 12:41 pm; edited 1 time in total

Hey guys, thanks for the links to the Gearslutz thread. Was a way more interesting discussion than myths about bit-depth/resolution. Learned loads from that thread, not so much from this thread

I feel like I know much more about digital audio than I did even last week.

Was I talking about compression except or MP3 as final medium? I was pretty sure I talked about gain staging and peak levels.





Hidden overs can occour if a signal is too hot (especially if peaked up until -1dBFS). What pretty much happens are spikes beyond 0dB (or glitches) that certain meters do not pick up. This is why there are meters that measure the so called "true peak", or have a measuring matrix like the old (now discontinued) InspectorXL. However you can apply certain measurement rules. Mostly they are "up to 3 samples over = no clipping" - but in best case scenario, you should have no overs at all.

They are not as present if you utilise a certain headroom (e.g.: -9dB according to the SMPTE recommendation, see below), but they are a major factor in the equation if you mostly work at 0dBFS digital.





I don't have my eyes everywhere, and I sure as hell have a different viewpoint about all this. Go to GearSlutz and hell would break loose. There is no exact right or wrong in terms of how to do things. But there are certain rules. Unfortunately they are mostly bent to their limits. I just wish people would get back to times before things went out of hands and "beyond the rulebook", with the benefits of a higher dynamic range (lower noise floor) due to modern ADCs/DACs. You can only do so much after all.





Actually, AFAICS, it's the other way around.

A practical example with Wavelab and it's internal BIT METERING tool (in WL5 days, this was the nice vertical GUI with up to 32blocks per channel and an additional red LED as warning light). If I have a file that is around K-20 (RMS of about -20dB at 600ms rise/fall) and the digital peak doesn't go higher than -5dB for example (which would be pretty normal in this case), the bitrate indicator can reaches up to full 24bit with 24bit recordings on transients, and about 10bit to 16bit at average level.

Judging by the current discussion - we "might lose" bitdepth and waste HDD space. But it can mean several things: plugins cap at at a certain bitrate, the DAC can only record at lower bitrates, if the bitrate metering shows bits below 24bit chances are you can truncate without dithering, you just have a healthy (and lively) level, etc.





Again I bring the hidden overs (is your DAW really save on it even on a per-channel basis?) into place on top of using actual outboard equipment. True, the file can be turned down prior to mastering, may it be OTB or purely ITB - heck I do it myself especially if I author a CD. But you still need to take the DAC into consideration, and at what worklevel your equipment is setup.

Here the feared -18dB (RMS) reference level comes to mind and the SMPTE recommendation to not exceed -9dBFS especially if you incorporate analog equipment and/or use QPPM tools (Quasi Peak Program Metering, read: 5ms rise, 300ms fall - see: Wikipedia). Also with the current rise of "analog type" plugins like Slate VCC, Waves, UAD, etc.

I bring yet again the example with a DAC of a CD player - one of mine barfs if there are (digital) peaks higher than -0,4dB (it's an old Phillips one). Trust me, I've tested a lot of CD players for compatibility with test-Masters back in the day and got the craziest results. On DAT I tend to keep away from higher levels than -1dB since I don't trust the ADC/DAC either.


Another example is at radio stations. Here these folks either ask for -6dB files, or "normalize" the files themselves down to that level. Reason is the MP2 format and the massive compression chain in the process. Too hot tracks will be squashed to nothingness regardless. This is one of the only ways to stir against it.

And while we're on CODECs: AC3 also has a loudness limit. I don't know details of hand anymore, but K-12 was too hot with tests back in the days. Depending on the used encoding matrix of course. MP3 starts to barf at -7dB RMS btw (tested wth a LAME encoder from 2011). Another reason why SONNOX Fraunhofer Pro-Codec exists.


To summarize:
I'd say proper gain staging is benefital and definitely not a waste of time. So is having a lively headroom.

I'm actually surprised that noone picked my "higher fader resolution" comment apart. Or did I even post it? Hm...

hidden overs is a curious but valid explanation.

I am pleased!

I would not have thought that something as arcane as gain-staging could generate such a contentious debate, but I have been wrong about such things before.

I would like to thank the audio engineers for their input. While I have to admit to being little more than a well-informed novice when it comes to audio production, I thought their posts were very useful. I can certainly appreciate their expertise.

My expertise is in computers. I know computers down to the processor level. I know what has to happen in the computer for all these wonderful things to happen in the DAW. The biggest myth I've heard in this thread is that if a DAW has 32-bit floating point for calculations, gain staging could not possibly make any difference.

I have been working with computers for a long time. One of the things young programmers had the hardest time getting their head around was that if 1 + 1 = 2, then why doesn't 1.0 + 1.0 = 2.0. I can't tell you how many bugs I came across because people didn't understand this. 1 + 1 is a integer calculation. 1.0 + 1.0 is a floating point calculation. 1.0 + 1.0 = 1.9999999.... Floating point works on approximations and uses significant digits. In other words, floating point calculations suffer from rounding errors. That's why 1.0 + 1.0 doesn't equal 2.0.

Now consider what is going on inside a DAW. There are hundreds of thousands of FP calculations going on every second. Small numbers produce small rounding errors and are inaudible. Large numbers produce larger rounding errors and they can become audible. If you insist on pushing levels to their max inside your DAW, you are going to lose information in the calculations.

Computers aren't magic. They have their own limitations. If you think you can do whatever you want as long as your output doesn't clip, you don't really know what is going on inside your computer.

Right. So this has nothing to do with gain-staging. What you're talking about is purely attacking 32 float processing period. You do enough processing of a float signal if you simply stack enough plugins together, you don't need to have some "poor gainstaging" ideology to accomplish that.

I have yet to find anyone who even CLAIMS to be able to hear the difference of two files where you supposedly have floating point error that reaches the level of audibility, even in some extreme case where you intentionally push the highest level possible into a series of plugins handling floating point.

actually in floating point the same rounding error is produced regardless of the magnitude. you have always at least 24 bits accuracy and if you happen to stay exactly where the exponent never changes you'll get 25.

consider this: if you work at +96db, converting the signal to 0db requires only changing the exponent. the data and the number of fractional bits you have remains identical!

when you're working with fixed point, smaller numbers produce higher rounding error. so even if you were right in some way (confusing float vs. fixed) you'd still have it backwards.

the "problem" with float is when you mix magnitudes. let's say you have an object with vertices at +/- 2^25. let's say you want to smoothly move the whole object by 1/x per frame. now you'll be throwing out so much data crossing that range of magnitude that you'll get no movement at all.

in other cases you get exploding objects which while really neat is pretty darn useless.

the solution? you accumulate the movement in one place, then adjust the magnitude, then apply it.

so you'd have:

object,
position,
velocity.

position += velocity;
displayed_object = object + position;

this prevents the issue from ever occurring. now the position of the object will be represented with at least 24 bits at 100% magnitude.

the opposite issue can occur too. if your position = 2^50, there will be no space left for the object and all points in the object will be equal to the same value.

one reason to take painfully care about levels is because the cpu process more accurately (without glitch for example) but this is rather important if you run a bunch of plugins simultaneously and want to use a low buffer with no surprise.

Guys, for f**k sake.

If you want to test this for yourselves, take a volume plugin, say ggain, drop the volume by 18db, put it at the beginning of a 5 or 6-stage effects chain (with no dynamics plugs like limiters, expanders, gates or compressors), then put the same plugin with +18db at the end of the chain.
Select both plugs and try playing back while flicking bypass on/off on them both.
Do the same thing but with +18db at the head of the chain and -18db at the end.

If you can't hear a difference, then no amount of info in this thread will help you. End of story.

You're welcome, JJ. It's a pleasure to share opinions with open minded people. I learn a lot here, and hope to give a little back.



No one is attacking anything here, rifftrax, except you and a few others irrational hatred of people with actual experience making records. Points can be made with respect or you can do it your way, repeatedly, more aggressively and louder.



I've helped friends and clients set up, record and mix on 32bit home DAWs too numerous and boring to go into here, but I work in the real world, where Pro Tools just happens to be an actual standard. Yeah, you can get away with more level using floating point, but why would you want to?

I DO understand 32 bit floating-point processing. I just record real people and things and don't think it's good practice to bury the meters in the red for any reason. Why would I want a vocal or guitar track that hot? As I mentioned earlier, real hardware like mic pres and compressors are at their worst in terms of signal to noise and distortion to create the levels you defend.

And turn something up just to turn it down again elsewhere? For dozens of tracks, sends, subs and master fader? Constantly readjusting your monitor levels? Not a great workflow. Meters don't matter at full code, no matter how much headroom you have. You might as well have signal on/off indicators. Ugh. And I don't make 30-100db miscalculations on my levels, so I really don't need it.

For the record, (literally) I just got home from tracking and rough mixing an album project for paying clients at an actual recording studio, so my e-penis DWARFS yours!

It's so big, it has snow on top in the summer! It's so big, it graduated a year ahead of me! It's so big, I was once in Ohio and got a blow job in Tennessee!

Xenobt

Wow, conceit much?



Yup, that's it. You nailed it. Your opinion on computers is actually worth more because you are the argument by authority. Brilliant. I see how this works now.

You do realize... how much you've just proven of my original point with your post right? The whole entitlement thing where because you make money from an artistic field it gives you license to be an authority on all kinds of pseudo-related topics? Natch.

Hey, whatever helps you sleep at night I guess.



Totally awesome - and real computers using real 32bit float processing certainly doesn't give two sh*ts about it. Is this thread about hardware gain-staging? No, it most certainly isn't. Maybe you forgot the point of the thread. Well, there it is.

Just tried this with a basic pop drum loop. Except I doubled the amount of gain to +36db to 'exacerbate' any "difference". Heard no difference whatsoever. I used a chain of Breverb, Fruity Parametric Eq, Fruity Chorus and Nomad Factory Liquid Delays 2. Exact same settings on both tracks. Zero discernible difference. For the record I have a pretty decent signal chain including a Saffire Pro 10, InterM R150 Plus reference power amp and HHB Circle 5 passive monitors.

Seriously, these rebuttals are getting straight up laughable. Did you even try this test you just proposed?

[edit] Well hell just to make DOUBLY sure I did the test as unforgivably "worst-case scenario" as possible (as also to point how incredibly f**king stupid these continued dumbass digital gain-staging arguments are) I went ahead and applied 90db of gain reduction and make-up flanking two DSP plugins on my test track while only having the two DSP plugins running with no gain-staging on the comparison track.

I kept Breverb (full wet percussive room with full dry signal) and some crazy mangled eq through FL studio's stock parametric eq on both tracks. One track had 90db of gain reduction preceding the two plugins and 90db of make-up gain immediately following it and one track had just the two main plugins. I had the vsti playing back the loop automated to switch between sending to each channel so the comparison would be immediate.

Sure enough. Zero detectable difference. Does anyone else want to continue to fight from the totally obnoxious wrong side of this argument? Would you like me to post audio examples?

[edit] Just did the opposite too. Introduced 90db of gain make-up and then the two plugs and then 90db of gain reduction. Zero difference/zero detectable distortion or anything. Case closed.

Moral of the Story:

Floating-point processing (32bit or above) does in fact mean that you can pretty much do whatever the f**k you well feel like in a digital gain-staging environment aside from special consideration for plugins that actually have level-dependent saturation/harmonics/controls/whatever, regardless of what some "actual engineer" might tell you. This includes having 100 decibel overs even before you hit your floating point processing plugin as long as your final level is below digital zero.

You too can fight ignorance by not blinding listening to computer advice given out by people who's only involvement with computers is mixing on them.

You're right if you're purely ITB - nothing should happen, you could even null the signal without noticable artifacts on a meter. If no soundshaping is involved, there should be none unless there is something wrong with your host.

But the main question is still in whether or not Gain Staging is a waste of time.

And I say not at all, especially on the long run. Not to mention if you want to share your projects (bad example, so do not do this: recordings over 0dB in 24bit integer will clip at 0dBFS, 32bit float will not!), incorporate hardware or use analog type (VST) plugins with an internal reference level of -18dB RMS. And they do exist for years now, not only since the "console wars" plugins.




There is a certain truth behind the sentence "don't touch the master fader".

I'm pretty sure you and I can agree on the fact that if we calibrated our work environment, leveled in the signal and use proper gain staging... we can totally ignore the master bus/fader, have a suitable fader resolution and still have a dynamic rage beyond 100dB in the digital realm.

Best of both worlds, no?




You should sell swag with that sentence on it. Didn't laugh that hard in a while (especially your following analogy). Well, except for that one mayhaps.

I learnt a bit about some technical stuff from you in this thread. You're obviously quite intelligent. A shame then that the arguments were wrapped in arrogance and disdain. In the above quote I bolded the relevant bit. Reading your posts in the light of this is quite ironic, as you seem to me to have a similar urge.

Not sure who wins the prize for the clever riposte pissing-match, you or rifftrax. Some people are impressed by such things I guess. Yourselves, I imagine.

I don't know why I carried on reading. I suppose it is quite fascinating watching people make such utter twats of themselves while they are so oblivious to it.

With each new post its like "LOOK WHAT AN ARSEHOLE I CAN BE NOW!!! ARE YOU IMPRESSED YET??? COZ I REALLY REALLY AM IMPRESSED WITH MYSELF!!!!"

sigh.
Last edited by someone called simon on Mon May 14, 2012 2:46 am; edited 1 time in total