Loud enough, never too loud

You know the moment: a programme is fine, then the break arrives and the advert is suddenly shouting. Nothing in the signal got taller in voltage. It got louder to your ears, and the gap between those two statements is the whole subject.

The oldest meter in audio measures the peak: the tallest single sample, how close it comes to the digital ceiling. It is essential for one job, making sure nothing clips, and close to useless for another, telling you how loud something will feel. Your ears do not respond to the tallest instant. They integrate energy over time and across the spectrum.

That mismatch is why two files can share an identical peak and feel worlds apart. A spiky drum hit can touch the ceiling and still feel quiet, because the loud part is over in milliseconds. A dense, compressed wall of sound can sit below the same ceiling and feel relentless, because it holds energy the whole time. For years the industry chased loudness by squashing everything toward the ceiling, the so-called loudness war, precisely because it was managing the wrong number.

The fix was to measure the way hearing works. LUFS, loudness units relative to full scale, averages the signal over time, but first it passes it through a weighting filter that mimics the ear. We are not equally sensitive across the spectrum: a tone at sixty hertz and a tone at three kilohertz at the same physical level do not sound equally loud, and our sensitivity itself shifts with volume. The weighting rolls off the very lows and gives a small lift through the presence region, so the meter counts energy roughly in proportion to how much the ear actually registers it.

The result is a number that tracks perceived loudness instead of electrical level. Two pieces matched in LUFS feel about as loud as each other, regardless of how their peaks compare or how aggressively either was compressed. The ruler finally measures the thing you care about.

Loudness depends on how long you measure, so the standard defines three windows for three questions. Integrated loudness averages an entire programme into one figure, the headline number a platform or broadcaster checks against. Short-term loudness uses a rolling three-second window, good for following how a mix breathes from scene to scene. Momentary loudness uses about four hundred milliseconds, fast enough to catch a sudden shout or impact.

They answer different things. Integrated tells you whether the whole piece sits at the right level. Short-term tells you whether it stays there or lurches. Momentary tells you what just spiked. A mix can be perfectly on target as an average and still feel uneven, which is why the average alone never tells the full story.

There is one more trap. Digital audio stores discrete samples, but playback reconstructs a smooth continuous wave through them, and that reconstructed wave can rise higher between two samples than either sample does on its own. A file that looks safely under the ceiling on a normal peak meter can overshoot once it is converted back to analogue or passed through a codec, and that overshoot distorts.

True-peak metering predicts those inter-sample peaks and reports the real maximum the playback chain will see. It is why delivery targets specify a true-peak ceiling a little below the absolute limit, commonly around one decibel of headroom, so there is room for the reconstruction and the encoder to do their work without clipping in someone's living room.

Once loudness could be measured the way it is heard, it could be agreed on. Broadcasters set integrated targets, the European standard around minus twenty-three LUFS, the equivalent North American figure near minus twenty-four. Streaming platforms normalise uploads toward their own reference, commonly in the region of minus fourteen, turning louder masters down rather than rewarding them. Making a track hotter no longer makes it win; it just gets turned back down to match.

That is the quiet end of the loudness war. Not a rule imposed from above, but a measurement that finally matched perception, so that the listener stops reaching for the volume knob between one piece of audio and the next. Loud enough to land, never so loud it has to be tamed on the other side.