Mixing Tips
Mixing Myths #03
Two years since the last of these, and the queue refilled without any help from me. Five claims, all of them out of my own inbox, all stated with the confidence of settled fact.
What makes this batch worth writing up is that none of the five is simply wrong. Every one has a real mechanism buried inside it, and in most cases the mechanism is more useful than the rule that grew out of it. Somebody once had an excellent reason to leave six decibels of headroom on a mix bus. The reason went away. The six decibels stayed, got repeated, and hardened into a number nobody I ask can source any more.
So this isn't debunking. It's separating the half that still works from the half that has only been handed down, because the first half changes how you mix and the second half costs you time.
“96 kHz sounds better than 44.1”
The true part first. A 44.1 kHz session carries everything up to 22.05 kHz, and there is no music up there. Whatever this argument is about, it was never about the recording.
It's about what your plugins do to the recording afterwards. Anything nonlinear manufactures content that wasn't in the source: a saturator, a clipper, a tube stage, a compressor with a fast detector. Feed a 5 kHz tone into one of them and the fifth harmonic wants to sit at 25 kHz. At 44.1 there is nowhere to put it, so it folds back down and arrives at 19.1 kHz instead, with no musical relationship to anything else in the track. That's aliasing, and it's a different animal from ordinary distortion. Harmonic distortion sits on top of the notes you played. Aliasing sits between them, which is why a little of it reads as grit rather than as warmth.
A real problem, then, with a local solution. Magnetite oversamples its own tape and amplifier stages and adjusts the factor to your session rate: more work at 44.1, less at 96, because less is needed up there. The VWB-1X puts the same thing on a switch, Off through 8x, so you spend the CPU on the one tube stage you're actually pushing instead of on all forty tracks at once.
The protection is already in the plugin. Raising the session rate mostly buys a second copy of it, at twice the CPU and a bit over twice the disk for the life of the project. Where the higher rate does pay is monitoring: a 128-sample buffer takes 2.9 ms at 44.1 kHz and 1.3 ms at 96, and a singer hearing themselves through a chain can feel that. Which makes it a tracking decision, not a mixing one.
“You have to mix in mono first”
Two different practices are wearing the same word here, and only one of them is compulsory. Confusing the two is how people end up doing the hard version of the easy one.
The compulsory one is the mono check, and it works because summing left and right is a brutally honest phase test. Anything sitting in both channels slightly out of time or out of polarity gets quieter or vanishes, and a wide stereo pad that was carrying half the arrangement turns out to have been carrying it on cancellation. The trap is the level jump. Material identical in both channels sums 6 dB louder; material that's uncorrelated sums only 3 dB louder. The centre of your mix leaps forward the moment you press the button, and depending on your DAW that button then drops the sum by 3 dB, by 6, or by nothing at all. Find out which yours does before you trust anything you heard in mono.
Mixing in mono as a working method is a different claim, and it comes from a world where a record had to survive an AM radio, one speaker in a shop, and a club rig somebody had wired in parallel. Those rooms still exist. But set your reverb amount, your panning and your width against a fold-down and you're compensating for a problem you can't hear yet. You will overshoot. Open it back up and it's too wide and too wet, reliably.
Thirty seconds in mono at low volume, listening for what disappears rather than for what sounds worse. Everything sounds worse in mono. That part isn't information.
“Leave 6 dB of headroom on the master bus”
I've asked a lot of people where the six comes from. Nobody has ever known, which is usually the sign of a rule that has outlived whatever it was protecting.
It was protecting something real once. Early digital consoles and fixed-point mix engines genuinely clipped at the bus, without warning and without recovery. Mastering engineers asked for peaks near −6 dBFS because they wanted room to work rather than a file already jammed against the ceiling. And there's a third reason that hasn't gone away: encoding to AAC or MP3 can push the reconstructed waveform above the sample peaks it was built from, so a master that measured exactly 0 dBFS comes back over the line on somebody's phone.
What has gone away is the clipping. Every DAW mixing natively sums in 32-bit float or better, and that format has roughly 770 dB of room above 0 dBFS. More to the point, it doesn't get coarser up there. The exponent moves, the 24-bit mantissa stays the same size, so a bus running at +9 dBFS internally isn't being damaged. It's a big number. Pull it down before the output and you get a bit-identical file to the one you'd have had by being careful.
The number survives, then, but its reason has moved. It counts wherever float meets fixed: the converter, a 24-bit render, a limiter's ceiling. And it counts for anything level-dependent, which on a mix bus is most of what you own. The VWB-1X calibrates −14 dBFS to 0 VU on its meter for exactly that reason. Run 6 dB hotter into it and you've broken nothing, you've just picked up a different piece of gear.
“Solo is how you find problems”
Solo is a good diagnostic tool that people keep using as a decision tool, and those are not the same job. It answers one question honestly and a different one misleadingly, and nothing in the button tells you which of the two you just asked.
The reason is masking. What a track contributes to a mix has little to do with how it sounds by itself, because everything around it is covering two thirds of its spectrum. A lead vocal that sounds thin and papery in solo is frequently thin on purpose: two guitars are holding 200 to 400 Hz and the vocal was staying out of the way. Solo it, hear the thinness, add 3 dB at 250 Hz to fix it, drop it back in. Now the vocal is buried in the one region where it had been clean. A correct decision, made against the wrong background.
Some moves can't be judged in solo at all. The VEQ-1P trick everyone knows, boosting and attenuating the same low shelf at once, works because the two filters are slightly off-tuned and leave a dip just above the boost. On a kick at 60 Hz in solo that reads as a mild, slightly odd loss of weight. In an arrangement it's what lets kick and bass share the bottom without eating each other.
Keep solo for faults you already suspect: a click, a chair creak, a bad crossfade, the frequency of a resonance you need to name. Otherwise mute. Pull the track out and hear what goes missing, which was the question you were asking.
“An A/B tells you whether the plugin helped”
It tells you which version was louder. Most of the time that turns out to be the same answer, which is exactly what makes it so hard to catch yourself doing.
Almost nothing you insert leaves the level alone. A compressor with makeup gain, a saturator, an EQ with a broad shelf, even a plugin sitting flat with transformers in the path. And the bias runs well below the point where you'd call anything loud. Two tenths of a decibel is under the level at which people reliably say “that one is louder”, and it is comfortably enough to decide which one they prefer. It doesn't arrive labelled as volume. It arrives as fuller, closer, more detailed, more finished, which is every word you'd use for a plugin doing its job.
Matching by peak won't save you, not with a compressor, because lowering peaks while raising the average is the whole function. Match peaks and the compressed version is still plainly louder. Match loudness instead, over the same eight bars, with short-term LUFS or an RMS meter and a gain plugin after the one you're testing, moved in steps of a tenth.
Then switch fast. Memory for timbre runs out in seconds, so a four-bar loop beats playing the song twice, and having somebody else do the clicking beats doing it yourself by a wide margin. Magnetite and the VEQ-1P both run as full 14-day trials. Set one up honestly and see whether you still want it when it isn't four tenths of a decibel louder.
Add a comment
You need to be logged in to add comments.