Broadcast loudness

Why did adverts sound louder than the programmes around them while meeting the same peak limit?

For decades the complaint was universal and the answer from broadcasters was, technically, correct: the adverts are not transmitted at a higher level than the programme. They were not. Regulations capped the peak level, the peak level was met, nothing was exceeded, and nobody was cheating.

Yet the perception was not an illusion either. The adverts really were louder in the only sense that matters to an ear. So we have a measurement saying two signals are equal and a room full of people saying they are not. When that happens, the interesting question is never who is lying. It is: what quantity is the instrument measuring, and is it the quantity anybody cared about?

Consider what a peak limit constrains. A transmitter has a hard ceiling — in analogue terms, the modulation depth beyond which the signal splashes into adjacent channels; in digital terms, the point where samples clip. The regulation says: never exceed this instantaneous value. That is a constraint on the single loudest moment in the signal, and on nothing else whatsoever.

Now ask what the ear does. Hearing does not respond to instantaneous samples; it integrates energy over some window — a few hundred milliseconds is the usual figure for the sensation of loudness to build. So perceived loudness tracks something close to the mean energy of the signal, not its extreme.

Put those two facts together and the gap opens by itself. Two signals can share an identical peak while having wildly different means. Define the crest factor as the ratio of peak to root-mean-square level. Natural speech and music have a high crest factor: brief transients far above a much quieter average. If you could reduce the crest factor — pull the peaks down relative to everything else — you would then have headroom, and you could apply gain to the whole signal until the peaks were back at the ceiling. Same peak. Much higher mean. Much louder.

Is there a device that does that? Dynamic range compression, followed by make-up gain and a limiter to catch what escapes — ordinary studio equipment doing what it was designed to do, applied deliberately.

Notice what has happened. Nothing illegal occurred. The rule was obeyed to the letter, and the rule was the wrong rule — it constrained an engineering quantity, set by transmitter physics, and was read by everyone as though it constrained a perceptual one. That mismatch is the entire mechanism.

The falsification test. This account says the extra loudness comes from a raised mean at an unchanged peak, and therefore from a compressed crest factor. The prediction is measurable: take an advert and the programme around it and compute both quantities. The account predicts equal peaks and a substantially higher mean for the advert, with the advert's dynamic range visibly narrower. The refuting observation would be an advert judged louder while measuring the same mean energy and the same crest factor as the programme — which would mean loudness was coming from somewhere else entirely, perhaps spectral balance or the density of speech. That is not what the measurements show; and the decisive practical confirmation came when regulators switched to a loudness measure and the disparity they had never been able to catch showed up immediately.

The fix is instructive because it had to be a measurement change before it could be a rules change. You cannot regulate a quantity you cannot measure, and until there was an agreed loudness meter, "don't be too loud" was unenforceable.

The ITU's recommendation BS.1770 supplied one: filter the signal with a weighting that approximates the ear's sensitivity across frequency, take the mean square over time, gate out the quiet passages so silence does not flatter the average, and report a single number in LKFS or LUFS. Broadcast targets were then written around it — around -23 LUFS in the European practice and -24 LKFS in the American, both recalled figures rather than derived ones, with a separate true-peak ceiling to keep the transmitter safe. Note the structure: a loudness target and a peak ceiling, because they are genuinely two different constraints and the old regime had only one of them.

Now separate the structural from the incentive, because both are doing work and they are usually blurred. The affordance is compression: the technology has existed for a long time and does not care what it is used on. The incentive is the advertiser's, and it is asymmetric — being noticed is worth more to them than being pleasant is. That asymmetry is why the escalation ran one way. And because loudness is judged relative to its neighbours, each participant raising their own level lowered everyone else's apparent level, which is the structure of a ratchet: nobody could stop unilaterally without appearing quieter. Blaming the compressor misses this entirely, and so does blaming individual advertisers — the same technology under symmetric incentives, as in a classical recording, is used to preserve dynamics rather than destroy them.

One belief worth naming as false: that adverts were transmitted at a higher volume. They were not, and the industry's denials on this point were accurate. What was misleading was the implication that being within the peak limit meant being no louder — a technically true statement deployed to answer a question it did not address.

• Whether the same normalisation on streaming platforms has actually ended the loudness war in recorded music, or merely moved it.
• How gating in the loudness measure prevents a quiet passage being used to buy headroom for a loud one.

Every term the collection defines is gathered in the glossary.