What is LUFS?

Loudness, not level

A peak meter reports the single highest instant in a signal, and that number says very little about how loud a whole programme feels. A voice compressed hard can sit near full scale from start to finish, while a quiet documentary with one door slam can peak at the same value. Integrated loudness measures the whole programme instead, so the figure describes what a listener actually hears. The EBU wrote its recommendation because peak normalisation had produced large loudness jumps between programmes and channels, and those jumps were the main source of audience complaints. Two companion figures usually travel with it: true peak, the highest point of the reconstructed waveform, and loudness range, which describes how much the level moves across the programme.

How the measurement works

A BS.1770 meter filters the audio to approximate human hearing, measures energy in short blocks, then gates out the quietest blocks so long pauses do not drag the average down. The result for a whole file is called integrated or programme loudness. Meters also show momentary and short term loudness, the second over a three second window, which help while mixing but are not the delivery figure. Because the scale is relative to digital full scale, almost every reading is negative: minus 16 LUFS is louder than minus 23. Moving from minus 23 to minus 16 is a 7 LU increase, the same as turning the entire mix up by 7 decibels.

The targets people deliver to

Broadcasters set a target and require it. EBU R 128 normalises every programme, adverts and promos included, to minus 23 LUFS, allows plus or minus 1 LU only where the target cannot practically be reached, such as a live programme, and caps true peak at minus 1 dBTP. The ATSC A/85 recommended practice sets minus 24 LKFS for American television. Apple asks podcasters to keep episodes around minus 16 LKFS, within 1 dB, with true peaks no higher than minus 1 dBFS. Many social and video platforms publish no figure at all, so the dependable habit is a consistent mix that never clips, because turning a file down later cannot repair distortion baked into it.

When it matters

Loudness only becomes visible when a video plays next to other audio. An advert inserted into a television break, a clip dropped into a playlist, a training module on an intranet beside twenty others: in each case a mix far louder or quieter than its neighbours makes viewers reach for the volume, and some of them leave instead. It matters inside a single video too. Speech has to sit clearly above music throughout, and a mix with a sensible integrated figure can still bury one quiet line, because the integrated number is an average of the whole file rather than a promise about any one moment.

Common mistakes

Normalising to peak and calling it finished is the most frequent, because a file with one loud transient ends up quiet overall. Chasing the loudest possible reading is the second: heavy limiting flattens speech, and a service that lowers loud files simply plays the squashed version more quietly. Measuring only a section, such as the first thirty seconds, gives a figure that does not describe the file. Ignoring true peak is the fourth, since lossy encoding can push the waveform above its original peak, which is why published ceilings sit below zero rather than at it.

Where it shows up in the product

The video builder here shows no LUFS figure and has no loudness target setting, so nothing in it normalises a finished video to minus 16, minus 23 or any other number. What it controls is balance. Background music starts at 25 percent volume, the slider stops at 60 percent, and wherever narration plays the music drops to 22 percent of its set level, ramping over 12 frames either side. The library holds 13 tracks, and your own audio file can be up to 25 MB. If a broadcaster, a podcast host or an internal platform names a loudness target, measure the exported MP4 with a BS.1770 meter and adjust it in an audio editor before delivery.

Questions people ask