Loudness, LUFS and why your station lurches

A station that lurches — a jingle that shouts, a track you have to lean towards, an advert that arrives like a slap — has one problem, and it is not mixing. It is measurement.

The files came from different places, made in different decades, mastered by different people to different targets. Nothing is wrong with any of them individually. Played back to back at face value they are unlistenable, and the listener's response is not to complain: it is to turn it off, because a station you keep adjusting the volume for is work.

Peak is not loudness

Two words get used as if they mean the same thing.

Peak is the single loudest instant in a file — how tall the waveform gets, measured in dBFS, with 0 dBFS as the ceiling nothing digital can exceed.

Loudness is how loud the whole thing feels while it plays.

A whisper with one door slam in it peaks high and feels quiet. A shouted advert, compressed until every syllable is the same size, feels enormous and may not peak any higher than that door slam.

This is why turning something up does not reliably make it louder. If a file already touches the ceiling there is nowhere for the peaks to go — push it and the tops are chopped off, which sounds harsh and cannot be undone. What you wanted was for it to *feel* louder, and that is a different measurement entirely.

What LUFS measures

LUFS — Loudness Units relative to Full Scale — is the broadcast standard for perceived loudness. It works the way hearing does rather than the way a waveform looks: frequencies the ear is most sensitive to count for more, and the measurement is integrated across the whole programme rather than taken at its loudest instant.

The numbers are negative and lower means quieter. A modern loud master might measure −7 LUFS. A 1970s pressing might measure −18. Both peak in the same place; one feels two and a half times louder.

Because it is a single number describing a whole file, LUFS is the thing you can actually correct with. Measure a track, compare it to a target, and you know precisely how much to turn it up or down for it to sit alongside everything else.

The loudness war, and why it is over

For about twenty-five years, mastering engineers competed to make records that felt louder than the ones next to them, on the reasonable assumption that louder got attention on the radio. The tools were compression and limiting, and the cost was dynamic range: the difference between the quiet parts and the loud parts, squeezed out until the waveform was a solid block.

What ended it was streaming. Every major platform now normalises playback to a loudness target, so a record mastered 6 dB louder than its neighbour is simply turned down 6 dB on the way to the listener. All the competitive loudness buys you is a record with less dynamic range than it should have, played at exactly the same level as everything else.

Radio automation is doing the same thing, for the same reason. Once playback is levelled, mastering loud stops being an advantage.

What a levelled station does

The mechanics are simple and unglamorous:

1. Measure on import. Every file is listened to end to end when it arrives, and its integrated loudness is stored alongside it.
2. Store a correction, not a change. The file is not re-encoded — nothing is lost, and a re-measurement is always possible. What is stored is an instruction: *play this 5 dB quieter*, *play this 2 dB louder*.
3. Apply it at playout. The correction happens when the track airs, so the loud master comes down, the quiet one comes up, and both sit at the same level.
4. Catch what is left. A limiter after the corrections holds back anything still trying to exceed the ceiling — files that arrived already clipped, or a mistyped number in a volume box.

The result is a library assembled from a dozen sources over thirty years that sounds like one station.

RadioHost Studio does exactly this, targeting −14 LUFS, which is where most streaming platforms sit and therefore where your listeners' ears are already calibrated.

The one thing to set by hand

Automatic measurement needs a file to measure, and there is one kind of audio that has no stable file: a bulletin fetched from somebody else's URL, replaced every hour. A measurement taken today describes yesterday's news.

Those get a manual offset in dB, set once by ear, applied to whatever arrives at that address from then on. It is worth setting while listening rather than guessing, and worth checking against the file's own headroom — pushing +6 dB into a bulletin that peaks at −2.9 dBFS clips it flat, and clipping is the one thing here that cannot be undone.

Two things people get wrong

"The station got quieter after I turned on levelling." It did, and that is levelling working. Most commercial libraries sit above target, so most files come down. The correct fix is at the very end — turn the encoder that feeds listeners up, once — rather than pushing individual files back up and undoing the evenness you just gained.

"I'll normalise the files before uploading." Peak normalisation, which is what most batch tools do, sets every file's loudest instant to the same height and does nothing for how loud they feel — the whisper-with-a-door-slam is still a whisper. You get an evenly-peaking library that still lurches, and you have re-encoded everything to achieve it.

Frequently asked questions

What LUFS should a radio station target?

−14 LUFS integrated is the common choice, matching where most streaming platforms normalise. The exact number matters far less than applying the same one to everything: consistency is the entire point.

Is LUFS the same as ReplayGain?

They solve the same problem and are not the same standard. ReplayGain came first, from the audio-player world; LUFS (as specified in ITU-R BS.1770 and EBU R 128) is the broadcast standard and is what modern tools measure. Both store a correction rather than changing the audio.

Will loudness levelling ruin the dynamics of my music?

No. Levelling adjusts the whole file by one amount — it turns things up and down, it does not compress them. What flattens dynamics is heavy compression or limiting, which is a separate decision and one you can decline.

Why does one track still stick out?

Its measurement probably predates the current audio — a replaced file keeps the old correction until it is imported again. Re-import it and it is measured afresh.

Do I need a limiter as well?

Yes, as a safety net rather than as a sound. Some files arrive already over the ceiling, and one mistyped number in a volume box is otherwise the difference between a small lift and a very unpleasant noise. On material that behaves, you never hear it work.