A single 30 Hz cycle takes 33 milliseconds to complete, over 300 times longer than one cycle at 10 kHz, and your limiter feels every bit of that.
That difference changes what a peak actually is. A snare crack pokes above your ceiling for a millisecond or two and is gone. A low bass note sits up there.
So when people talk about taming peaks in a master, they are usually picturing spikes. Half the time the real culprit is a note, held, doing damage for far longer than any drum hit ever will.
Set the release fast enough to track that, and the gain reduction starts modulating inside the waveform itself. That is not level control anymore. That is distortion, and it shows up as a gritty edge on the bass that nobody asked for.
Set the release slow, and you get the opposite problem. The whole mix ducks, holds, and comes back up around every low note, which is the pumping most people blame on the limiter when the source was really the arrangement.
Meanwhile, your ceiling is fixed. Spotify normalizes tracks to -14 LUFS and asks for masters below -1 dBTP, dropping that to -2 dBTP when the master is louder than the target, because lossy encoding pushes inter-sample levels higher than your file suggests. The AES recommendation for streaming lands on the same -1 dBTP figure at the codec input. Neither number is a suggestion you can safely ignore, since exceeding it moves the distortion downstream into the encoder, where you have no control over it.
The cost is phase. Every minimum-phase filter smears timing around its cutoff, and if you set that cutoff high enough to meaningfully reduce a 45 Hz peak, you have also softened the front edge of the kick. Linear-phase versions trade that for pre-ringing, which on low frequencies is long enough to hear as a soft blur ahead of the kick.
Folding the low end to mono is the other standard move. It tightens things up and helps on vinyl, though it does very little to the peak level itself, because summing two correlated channels does not lower amplitude.
Compressing the bass before the limiter is closer to the right idea. Closer, but still global. It still applies its curve to every note in the track, including all the ones already sitting exactly where you wanted them.
That is the argument for gain envelopes drawn ahead of the limiter, which flatten only the offending swell and leave the rest of the performance untouched. The notes that overshoot get pulled down. Nothing else moves.
When those two are roughly in phase, the combined peak can exceed either source on its own by a wide margin. When they are out of phase, the same collision produces a dip, which is why the problem shows up on some notes and not others inside a single track.
This is trivially fixable in the mix. Move the note, retune the kick, sidechain the bass. At the mastering stage, those options are gone, and you are working with a stereo file where the two are permanently summed.
What you can still do is find them. Zoom into the waveform of a finished mix, and the long, sustained low-end swells that break the ceiling are visible and countable, which makes them targetable one at a time.
That distinction matters more than it sounds. A global process has to stay conservative because it applies everywhere, while a fix aimed at a handful of specific moments can afford to be aggressive, since it only touches those moments.
It also explains why perceived loudness barely drops. Equal-loudness behavior means low frequencies need considerably more level than midrange to register as equally loud, so pulling a couple of decibels off a sub-bass swell costs you almost nothing in how loud the track feels, while handing real headroom back to everything above it.
The number to watch is not average gain reduction. It is how often per song your limiter drops into deep reduction, and how long it stays down each time.
That difference changes what a peak actually is. A snare crack pokes above your ceiling for a millisecond or two and is gone. A low bass note sits up there.
So when people talk about taming peaks in a master, they are usually picturing spikes. Half the time the real culprit is a note, held, doing damage for far longer than any drum hit ever will.
Your limiter reads a bass note as one sustained event
Peak detection does not care about frequency. It cares about amplitude over time, and a sustained 40 Hz swell hands it 25 milliseconds of continuous work per cycle.Set the release fast enough to track that, and the gain reduction starts modulating inside the waveform itself. That is not level control anymore. That is distortion, and it shows up as a gritty edge on the bass that nobody asked for.
Set the release slow, and you get the opposite problem. The whole mix ducks, holds, and comes back up around every low note, which is the pumping most people blame on the limiter when the source was really the arrangement.
Meanwhile, your ceiling is fixed. Spotify normalizes tracks to -14 LUFS and asks for masters below -1 dBTP, dropping that to -2 dBTP when the master is louder than the target, because lossy encoding pushes inter-sample levels higher than your file suggests. The AES recommendation for streaming lands on the same -1 dBTP figure at the codec input. Neither number is a suggestion you can safely ignore, since exceeding it moves the distortion downstream into the encoder, where you have no control over it.
High-passing is a blunt fix for low-end distortion
The usual first move is a high-pass filter, and it does work, up to a point. Below about 30 Hz, there is rarely anything you want to keep on a pop or rock master.The cost is phase. Every minimum-phase filter smears timing around its cutoff, and if you set that cutoff high enough to meaningfully reduce a 45 Hz peak, you have also softened the front edge of the kick. Linear-phase versions trade that for pre-ringing, which on low frequencies is long enough to hear as a soft blur ahead of the kick.
Folding the low end to mono is the other standard move. It tightens things up and helps on vinyl, though it does very little to the peak level itself, because summing two correlated channels does not lower amplitude.
Compressing the bass before the limiter is closer to the right idea. Closer, but still global. It still applies its curve to every note in the track, including all the ones already sitting exactly where you wanted them.
That is the argument for gain envelopes drawn ahead of the limiter, which flatten only the offending swell and leave the rest of the performance untouched. The notes that overshoot get pulled down. Nothing else moves.
Kick and bass fight over the same 30 milliseconds
Most rogue low-frequency peaks are not one instrument. They are a kick and a bass note landing close enough in time and frequency that their waveforms sum.When those two are roughly in phase, the combined peak can exceed either source on its own by a wide margin. When they are out of phase, the same collision produces a dip, which is why the problem shows up on some notes and not others inside a single track.
This is trivially fixable in the mix. Move the note, retune the kick, sidechain the bass. At the mastering stage, those options are gone, and you are working with a stereo file where the two are permanently summed.
What you can still do is find them. Zoom into the waveform of a finished mix, and the long, sustained low-end swells that break the ceiling are visible and countable, which makes them targetable one at a time.
That distinction matters more than it sounds. A global process has to stay conservative because it applies everywhere, while a fix aimed at a handful of specific moments can afford to be aggressive, since it only touches those moments.
It also explains why perceived loudness barely drops. Equal-loudness behavior means low frequencies need considerably more level than midrange to register as equally loud, so pulling a couple of decibels off a sub-bass swell costs you almost nothing in how loud the track feels, while handing real headroom back to everything above it.
The number to watch is not average gain reduction. It is how often per song your limiter drops into deep reduction, and how long it stays down each time.
