Your ears resolve a level difference between them of about 1 dB, which is finer than most people assume, and finer than many unlinked processors stay within.
That number matters the moment a processor treats your two channels as separate problems. Pull 2 dB off the left channel while the right one sits untouched, and you have manufactured a 2 dB level difference that was not in the mix.
The catch is that it does not last. The imbalance exists only while the gain reduction is active, so it appears, moves something, and vanishes before you can point at it.
For scale, the largest timing difference a human head produces naturally is somewhere between 660 and 790 microseconds, and listeners can resolve horizontal position to within one or two degrees straight ahead. Your hearing is not approximate about this.
Level differences are not equally useful at all frequencies, though, and that detail changes everything about how an unlinked gain move sounds. Below about 1 kHz, level difference is close to useless as a directional cue. Above 4 kHz, it becomes a strong one. Between those two figures, the cue fades in gradually.
So when one channel dips on a snare hit, the bass stays exactly where it was and the cymbals, consonants, and room tail slide sideways. That is why the effect reads as smearing rather than panning. The low end anchors, the top end drifts.
That extra reduction is what people hear as narrower and flatter. An unlinked processor gives each channel its own dynamics back, which is why unlinked settings are so often described as wider and more open.
Both descriptions are accurate, which is the frustrating part. Working mastering engineers tend to land on the same practical rule, that a couple of decibels of unlinked reduction causes no trouble, and the risk climbs the harder you push.
Which is why the amount of stereo linking you dial in is usually the wrong control to obsess over. Depth is the variable that decides whether unlinking is safe, and depth is set by how far above threshold your loudest moments sit.
That points at taking the overshoots down before the link setting matters at all. Flatten the few moments that force deep reduction, and the processor after them never goes deep enough for the channels to diverge audibly.
When the two channels are momentarily unequal, the mono sum of that moment is not the mono sum of the surrounding material. Content that was centered no longer sums the way centered content should, and the balance tilts for as long as the gain move lasts.
Pan law explains why this is worse than it looks. Under linear panning, a centered signal already arrives 6 dB down in each channel before summing, so any per-channel gain change lands on material that was already relying on both sides being equal.
The test is simple enough to do on every master. Fold to mono, loop the loudest section, and listen for the moments where the center of the image steps to one side and steps back.
You are not listening for width. Width is easy to hear and rarely the problem. Nobody complains that a master is too wide. You are listening for whether the middle of the record stays where you put it when the loud parts arrive.
A lead vocal that wanders a few degrees left on every chorus downbeat will not show up on any meter you own. It shows up as a record that feels unsteady, and nobody will be able to tell you why.
That number matters the moment a processor treats your two channels as separate problems. Pull 2 dB off the left channel while the right one sits untouched, and you have manufactured a 2 dB level difference that was not in the mix.
The catch is that it does not last. The imbalance exists only while the gain reduction is active, so it appears, moves something, and vanishes before you can point at it.
One decibel is enough for your ears to notice
Human localization runs on two cues. Timing differences between the ears, where the smallest detectable gap is around 10 microseconds, and level differences, where the threshold sits near 1 dB.For scale, the largest timing difference a human head produces naturally is somewhere between 660 and 790 microseconds, and listeners can resolve horizontal position to within one or two degrees straight ahead. Your hearing is not approximate about this.
Level differences are not equally useful at all frequencies, though, and that detail changes everything about how an unlinked gain move sounds. Below about 1 kHz, level difference is close to useless as a directional cue. Above 4 kHz, it becomes a strong one. Between those two figures, the cue fades in gradually.
So when one channel dips on a snare hit, the bass stays exactly where it was and the cymbals, consonants, and room tail slide sideways. That is why the effect reads as smearing rather than panning. The low end anchors, the top end drifts.
Linking trades a wider image for a stable one
A linked processor lets the louder channel drive both. Relative level between left and right is preserved exactly, so nothing moves, but both channels get reduced whenever either one asks for it.That extra reduction is what people hear as narrower and flatter. An unlinked processor gives each channel its own dynamics back, which is why unlinked settings are so often described as wider and more open.
Both descriptions are accurate, which is the frustrating part. Working mastering engineers tend to land on the same practical rule, that a couple of decibels of unlinked reduction causes no trouble, and the risk climbs the harder you push.
Which is why the amount of stereo linking you dial in is usually the wrong control to obsess over. Depth is the variable that decides whether unlinking is safe, and depth is set by how far above threshold your loudest moments sit.
That points at taking the overshoots down before the link setting matters at all. Flatten the few moments that force deep reduction, and the processor after them never goes deep enough for the channels to diverge audibly.
Mono fold-down is where the damage shows up
Stereo playback forgives a lot. Mono does not, and mono is still what you get from a phone speaker, a kitchen radio, a club system with the subs summed, and most laptop playback. Phone speakers in particular reproduce almost none of the low end, which leaves exactly the frequency range where level difference dominates localization.When the two channels are momentarily unequal, the mono sum of that moment is not the mono sum of the surrounding material. Content that was centered no longer sums the way centered content should, and the balance tilts for as long as the gain move lasts.
Pan law explains why this is worse than it looks. Under linear panning, a centered signal already arrives 6 dB down in each channel before summing, so any per-channel gain change lands on material that was already relying on both sides being equal.
The test is simple enough to do on every master. Fold to mono, loop the loudest section, and listen for the moments where the center of the image steps to one side and steps back.
You are not listening for width. Width is easy to hear and rarely the problem. Nobody complains that a master is too wide. You are listening for whether the middle of the record stays where you put it when the loud parts arrive.
A lead vocal that wanders a few degrees left on every chorus downbeat will not show up on any meter you own. It shows up as a record that feels unsteady, and nobody will be able to tell you why.
