Three ways to make a peak smaller

Why the loudest-sounding masters are usually the ones where the limiter did the least.

Near the end of the chain there are three stages in a row that, described arithmetically, do the identical thing: they reduce the distance between the loudest moment and the average. Saturation, then the limiter, then a safety clipper.

Three tools, one job. And the choice of which one does most of the work is probably the single biggest factor in whether a master sounds expensive or exhausting.

Why peak reduction is the whole game

Loudness normalisation is going to set your master to a fixed level — -14 LUFS or wherever the platform wants it. That number is measured on the average, not the peak.

So the only way to make a master sound bigger at a fixed average is to get the average closer to the peak. Every stage below is a different answer to that question, and they differ entirely in what they cost.

Saturation: buying level with harmonics

Saturation is a deliberate non-linearity. Feed a signal in, and instead of coming out proportionally larger, the loudest parts come out slightly less large than they went in — the transfer curve bends.

Two things happen at once. The peaks get rounded, which is the arithmetic. And the bending generates new frequency content that was not in the original — mostly harmonics of what is already there, so it is musically related rather than random.

That second part is why saturation is not just quiet limiting. Adding harmonics above a fundamental makes it sound louder and brighter without being louder, because your ear judges presence partly from harmonic content. A bass note with added second and third harmonics reads as bigger on a laptop speaker that cannot reproduce the fundamental at all.

The two controls do different jobs:

• ·drive is how hard the signal is pushed into the bend. More drive, more harmonics, more peak reduction, and eventually audible distortion.
• ·mix is how much of the saturated signal is blended back with the clean one. At 0.3 you are hearing mostly the original with the harmonics added underneath.

The mix control is what makes saturation safe. Parallel blending means you get the added density without losing the clean transient entirely — the untouched signal is still in there at full height.

The limiter: buying level with gain reduction

A limiter is a compressor with an infinite ratio and look-ahead. It reads a few milliseconds into the future, sees a peak coming, and turns the gain down just enough and just early enough that the peak arrives exactly at the ceiling.

It is designed to be inaudible, and within reason it is. The cost is that it is doing the same thing as a compressor with a very fast attack: removing the top of transients. A decibel or two of this and you hear nothing. Six decibels and the master is flat, dull and tiring, and no EQ afterwards will restore it, because what was removed was information.

Two settings matter:

• ·ceiling is the destination — where peaks are allowed to reach. -1.0 dB is the usual choice, and the headroom below zero is not caution: lossy encoding reconstructs a waveform that can overshoot the samples it was built from, so a file that peaks at exactly 0 dBFS will clip after streaming compresses it.
• ·release is how quickly gain recovers. Too short and the gain is moving audibly within single cycles of a bass note, which is distortion. Too long and a loud moment pulls the level down for the passage that follows, which is pumping. 100 ms is a middle that suits most material and suits none perfectly.

The clipper: the one that isn't a tool

The safety clipper sits after the limiter, at -0.3 dB, and this order is deliberate:

After the limiter on purpose — it exists to catch what the limiter's look-ahead did not, so putting it earlier would leave the last peaks unguarded.

A clipper does not reduce gain. It simply refuses to output anything above its ceiling — the waveform gets a flat top. That is distortion, unambiguously, and on sustained material it sounds like it.

Which is why it is set where it is: below the limiter's ceiling by a fraction, so that in normal operation it never engages at all. It is not there for character. It is there because look-ahead limiters occasionally miss a sample, and a master that overshoots by one sample is a master with a defect in it.

Some engineers use clipping deliberately and aggressively, especially in loud genres, because a flat-topped transient is enormously efficient at raising average level and a two-millisecond flat top on a snare is genuinely hard to hear. That is a legitimate technique. It is not what this stage is for.

The combination is the real answer

Here is the point all three stages are building towards.

Take a mix that needs 6 dB of peak reduction to reach its target. Do all of it with the limiter and you get 6 dB of transient removal: loud, flat, fatiguing.

Now saturate gently first. The saturation removes 2 dB of peak while adding harmonic density that makes the result read as louder still. The limiter now only has to find 4 dB. Same final loudness, noticeably more life.

Almost every master that sounds crushed is one where the limiter was asked to do work that an earlier stage should have done. Not because the limiter is bad — because it is the only stage in the chain with no way to be gentle. Its ratio is infinite by definition. Everything upstream has a dial that lets it do a little; the limiter only decides how much.

This is also why the chain compresses twice before it gets here. Every decibel of peak that the glue and tonal compressors handle musically is a decibel the limiter does not have to handle brutally.

And none of it makes the master louder

Worth restating, because it undoes the intuition that all of this is a loudness competition: the stage after the clipper normalises to a target. Push the saturation harder and you do not get a louder master — you get a differently textured master at the same loudness.

Which is a much better deal than it sounds. It means these controls are free to be used for what they are actually good at, and the number at the end takes care of itself. What you are choosing is not level. It is what the record costs your ears over forty minutes.