Controlling your peaks

I've read that when mixing, you don't want any of your meters to go over 0dB. I know that when recording, going over 0dB will result in clipping distortion which sounds terrible. However, when working with VSTi instruments, I can go as far as I want over 0dB and not have any distortion. So is there any reason to keep things below 0dB in this case, especially since I'm probably going to end up limiting things in the end?
posted by god particle (7 comments total)

In most computer workstation software (virtually all the majors), there is "infinite" headroom in the internal signal path, it neither clips nor limits these signals.

It is not unheard of for me to have a track inside Reason (a synth software) or SONAR (multitrack audio and synth SW) that actually goes up to +18db in transient peaks. But watch out for the final output -- keep those below zero. The faders either always end up below -18db by the time I finish my mix, or have a compressor/limiter applied with 0ms (or even 4ms "lookahead") attack.

So long as you don't hear any distortion whatever is above 0dB is probably still in the "internal signal path," so you're fine.
posted by chimaera at 4:57 PM on July 6, 2008

Yeah, exactly for chimaera said. It's okay to let it go to 0dB+ for very small intervals too, and will sometimes produce a bit more punch in the sound.

Just, you know, don't compress the hell out of it.
posted by spiderskull at 9:29 PM on July 6, 2008

Just, you know, don't compress the hell out of it.

And if you do, post it here, get your feelings hurt (but just a little), and then get back to work mixing and remixing until you get a decent balance of desirable compression characteristics and punch.

Also, be sure to apply the limiter at the post-fader stage in the signal path; it doesn't do much good if you apply it pre-fader. And as I recently re-learned the hard way on a couple of rough mixes, be sure not to lean too hard on the limiter (that is, watch that the gain on the input signal to the limiter isn't just totally off-the-charts to start with or you'll end up with an over-compressed signal like spiderskull warned against). And this last point is especially important if you're routing a bunch of individual channels that already have some limiting applied to them into a sub-mix in a group channel.

And then after all that, don't over-compress the stereo mix either when/if you master it, or you can bring out compression artifacts that might not have been noticeable otherwise.
posted by saulgoodman at 6:54 PM on July 8, 2008

I mildly and respectfully disagree. I believe that Cubase and Nuendo, for instance, internally maintain a 32 bit floating point word length, as opposed to the 24 or 16 bit integers of your actual PCM (wav/aiff) files. This, of course, makes for a much larger dynamic range.

In this sense, what's labelled "0 dB" on your DAW software's channel faders is really much lower than the actual glass ceiling of zero dBFS. I presume they do this to mimic the behaviour of analogue mixers, which tend to have some headroom above 0 dB ("unity gain") as well.

Now, when digital recording was starting to become commonplace in the nineties (think ADATs and the early adopter DAWs), the 16-bit limitation and crappy, crappy AD/DA converters caused people to shoot for the sky level-wise, calibrating 0 dB analogue at -10 dbFS digital or even higher, taking the (not so) rare overload light in stride, or just doing another take when it clipped. These days, with 24 bits to spend (remember, this is a theoretical increase of 48 dB over 16 bits) and decent converters in even the cheapest gear, noise floors and low-level quantizing crud are much, much less of a concern.

Thus, it is considered good practice to keep unity gain in the analogue domain calibrated to a fixed figure in the digital, with 0dB analogue = -18dbFS digital being the most commonly cited equivalency.

I happen to be quite lazy and absent-minded and I am not very orthodox in observing these standards. The consequence of this is that I tend to forget about them, resulting in the occasional ruined take. So, I try at least to keep in mind to maintain a nominal level of around -20 dBFS when recording, peaking at -12 at most. In rock & roll, this holds especially true for drums and vocals, which can see peaks jumping up to 10 dB over what you had at sound-check easily in the heat of a spirited take.

Now, you may not record a lot of analogue sources, or any - but aside from the laziness/confusion argument it should be noted that if you stay in the box, it holds especially true that there isn't a lot to worry about in the way of converters and noise floors (since everything is digital from start to finish and there's no AD/DA stage). In modern digital, it really costs nothing to stay well below unity, so why not do it?

Lastly, even if any given DAW application offers channel headroom well over (nominal) "unity", the master buss will probably be much less forgiving. I know that in Cubase SX 3, whereas headroom is copious on channels and busses, 0dB on the stereo master buss really is 0dBFS, no prisoners.

This means that in a busy mix with a lot of channels running quite hot, one can count on "level creep" (a common experience during mixdown, for me at least) causing the mix to eventually clip the master buss, forcing you to pull down the levels on all tracks equally, which absolutely takes the urgency out of the mix process for me. (And even so, reducing the pre-fader trim/gain levels on all channels equally tends not to work, because you might have some plugins in the insert channels which will behave differently when offered a lower-gain input signal.) Keeping things less hot from the get-go helps counteract this, for me.

Not to mention that if you plan on moving your stereo (or surround) mix on to post-processing or mastering at all, it is generally rewarding to maintain some headroom on even *that* - I try not to go over -4 dbFS peak on an unmastered stereo mix, myself. You or the mastering engineer will simply want some headroom to work with. If you offer a mix that peaks close to full scale to a mastering house, the first thing the engineer may well do is lower the volume by several dBs. Not only is this principally unnecessary work you might be paying them good money for, it's also true that any destructive manipulation of a digital sound file (i.e. gain or most effects except reversal, channel flip or anything else that can be restored to the original bits 1:1 post-hoc disregarding internal undo histories) causes a slight degradation of the signal, although again this is much less of a problem with contemporary technology, and frankly, this veers well into total geek territory.

Bottom line: you probably have nothing to lose. So why not stay cool on those files? You can always turn up the monitors, and you'll have tons of headroom for mixing to boot. Just remember to turn them well down before playing a commercial recording (which will be much louder) - you might break your ears, or those of your neighbours.
posted by goodnewsfortheinsane at 7:32 PM on July 8, 2008 [2 favorites]

Er, this is in response to the first two comments, I'm only reading Saul's just now.
posted by goodnewsfortheinsane at 7:34 PM on July 8, 2008

I think we're in violent agreement, gnfti. Regardless of what the individual tracks are doing, what ultimately matters in computer-based recording is what's going on at the master output.

I, personally never, ever, ever let the master above 0db in the final mix, but some individual tracks peak above zero fairly regularly. On (very rare) occasion I may push to a little digital clipping during mastering but let your mastering engineer make that call.
posted by chimaera at 9:13 AM on July 9, 2008

I just rendered a song after pretty much ignoring the fader levels to see if they really made a difference when working exclusively with software, and the song did end up having quite a bit of clipping, so from now on I'll try keeping things under 0dB.
posted by god particle at 10:53 AM on July 9, 2008