Why these six
A mix engineer can change level, tone, dynamics and space. What they cannot change is information the recording never captured, or damage that overwrote it. Every item below is in the second category, which is why they are worth the ten minutes they cost at the microphone.
The list is deliberately short. Recording advice tends to run to fifty items, most of which are preferences — microphone choice, preamp, cable, the room’s reverb time — and preferences are recoverable, because a different one produces a different recording rather than a broken one. These six are the ones where the take is worse in a way no later decision reaches.
1. Recording into the red
Digital clipping is not a warm saturation, it is the top of the waveform replaced with a flat line, and the samples that described the peak are gone. No plug-in reconstructs them — the best restoration can do is guess a curve, and a guessed curve on a vocal consonant or a snare transient is audible as a guess.
There is also no reason to risk it. At 24-bit, peaking around −12 dBFS leaves enormous headroom and costs nothing in noise; the “record as hot as possible” habit is inherited from tape and 16-bit, where it made sense and no longer does. The noise floor of a 24-bit converter sits far below anything a home room adds, so the level you record at has no bearing on how clean the take is. It only decides whether the loudest note survives.
Set the gain on the loudest thing the performer will do — the shouted chorus, the hardest hit — not on the sound check, which is always quieter. And watch the interface’s meter rather than the DAW’s: clipping happens at the converter, and a track that shows −3 dBFS in the session can already be flat-topped if the preamp was overdriven before it. If you suspect a take is already damaged, the clipping detector finds the flat-topped runs and tells you where they start, in the browser, without uploading anything.
2. Recording the room instead of the source
A microphone a metre from a singer in an untreated bedroom captures the singer and the room in roughly equal measure. Reverb can be added later; the room already on the recording cannot be removed, because it arrives at the same time as the thing you wanted and is made of the same frequencies.
The fix is distance and soft material, not equipment. Closer to the source, further from parallel walls, something absorbent behind the microphone and behind the singer. A duvet genuinely helps and a better microphone genuinely does not — a more sensitive microphone hears the room more faithfully, which is the opposite of the aim.
The test is cheap: record ten seconds of a hand-clap where the singer will stand, and listen for the flutter and the ring after it. If the clap has a tail, so will every word, and the tail is the part that turns to mud when the vocal is compressed later.
3. Leaving a noise floor you will want gone
Computer fans, air conditioning, traffic, a fridge two rooms away. Individually inaudible on one track; across twenty tracks all summed together, an audible haze that compression then pulls up between phrases.
Noise reduction is one of the few things that genuinely improves with modern tools, but all of them trade artefacts for silence, and the trade gets worse the more you ask. A steady broadband hiss at a low level is the best case and is usually recoverable; a fan whose pitch wanders, or a noise that sits in the same band as the voice, is the worst. Turn things off before recording rather than after. The same free check reads the noise floor, which is the number to know before deciding whether a take is worth keeping.
Record ten seconds of the room with nobody in it, at the take’s gain, and listen back loud. Whatever you hear is on every take. That same ten seconds is also the sample any restoration tool will want later, so keep it either way.
4. Fixing timing by nudging every take
Editing a performance onto a grid removes the thing that made it feel like a performance. Ensembles do not play exactly together — the small, consistent relationships between players are what a listener hears as groove, and quantising erases them uniformly.
The mistake is not editing; it is editing instead of performing. A take with one late entrance is fixed by moving one entrance. A take that needs every note moved is a take that should be recorded again, and the difference between the two is the difference between a repair and a replacement. The same logic governs pitch — tuning is for the four or five notes that are wrong, not for every note in the take.
This one is reversible in principle and never in practice, because nobody keeps the unedited take. Keep it. A duplicate of the raw playlist costs nothing, and the mix that wants the human timing back two weeks later is common.
5. Two microphones on one source, unchecked
Two microphones at different distances from the same source capture the same waveform at different times. Summed, some frequencies reinforce and others cancel, and the result is a hollow, filtered sound that no equaliser fixes — because the missing frequencies were cancelled, not attenuated. There is nothing to boost.
It happens more often than the phrase “two microphones” suggests: a close microphone and a room microphone on a guitar amp, a top and bottom microphone on a snare, a vocal microphone and the headphone bleed of the same vocal, a direct box and a microphone on the same bass cabinet.
Check it at the source: sum both microphones to mono and move one until the low end is fullest, or flip the polarity on one and keep whichever is fuller. Thirty seconds then, unfixable later. On a finished file the same problem shows up as negative correlation between the channels, which is what the stereo width analyser reads per band — a useful second opinion when a mix sounds thin in mono and you cannot hear why.
6. Mixing on one pair of speakers, loud
Every room lies about the low end and every listener stops hearing the top after an hour at volume. A mix made under those conditions is corrected for a problem that only exists there: the bass that was turned down because the room boomed, the top end that was pushed because tired ears had stopped hearing it.
Not a recording mistake exactly, but it is the one that produces the most unfixable results — because a mix balanced wrongly on purpose is much harder to diagnose later than one that was simply never balanced. Check on headphones, on a phone speaker and in a car, at a level where you could hold a conversation, and take a break before every decision that matters.
What can still be fixed later
Being honest about the six above means being equally honest about the rest, which is most of it. Level, tone, dynamics, space and arrangement balance are all decisions that can be revisited for as long as the separate tracks exist, and a mix that is merely unbalanced is a mix, not a problem.
- Steady noise and moderate room are the recoverable end of items two and three. Separating the defect from the source and treating each on its own is what audio restoration does, and the guide to removing noise covers which kind of noise you have before reaching for it. Neither can give back what a wandering fan or a hard flutter echo took.
- Balance and masking — the vocal buried under the guitars, the bass and the kick fighting — are mix problems, and the reason the six above matter is that they are the ones a mix cannot reach. AI mixing works from the separate stems and produces a starting balance with its reasoning listed, which is the useful half of a second opinion.
- A single late note or a slightly flat one is repairable without flattening the performance, which is the whole distinction item four draws. The line is between fixing a moment and rebuilding a take.
Ten minutes before the red light
- Gain set on the loudest thing the performer will do, peaking around −12 dBFS.
- Ten seconds of room tone recorded and listened to, loud.
- A hand-clap where the source will be, checked for ring and flutter.
- Everything with a fan or a compressor motor switched off, or moved.
- Every pair of microphones on one source summed to mono and checked for a hollow low end.
- A copy of the raw playlist kept before any edit.
The pattern
Every item here is a decision made before the recording exists, and the reason they stay expensive is that processing works on what was captured rather than on what happened. Almost everything else — level, tone, space, arrangement balance — genuinely is fixable later, so it is worth spending the attention where it does not come back. What that attention buys is described in mixing vs mastering: a mix that only has to balance the parts, because none of them arrived broken.
Keep reading