Open a file in an editor and the meter shows how close the loudest sample came to the ceiling. That number is easy to read and answers almost nothing a delivery spec cares about. Two files can both peak at −3 dBFS: one a quiet interview with a single door slam, the other a wall of compressed music with no dynamics at all. They will not sound remotely alike, and only one of them will pass.
Loudness specifications exist because peak level and perceived loudness came apart the moment broadcasters started normalising by peak. Everyone compressed harder to get closer to the ceiling, adverts got louder than programmes, and viewers reached for the remote between every segment. The fix was to measure something closer to what an ear does.
What LUFS is actually measuring
LUFS stands for Loudness Units relative to Full Scale. The procedure behind it is ITU-R BS.1770, and it is worth knowing roughly what it does, because the details explain most of the surprises.
The signal is first pushed through two filters, together called K-weighting. The first is a high shelf that adds about 4 dB above 2 kHz, standing in for the way a head and torso affect what reaches the ear. The second is a high-pass that rolls off the bottom end, because low-frequency energy contributes far less to perceived loudness than its raw power suggests. A bass-heavy mix and a dialogue mix with the same RMS level do not sound equally loud, and this is where that gets accounted for.
After filtering, the mean square is taken per channel and the channels are summed with weights. Left, right and centre count fully. The surrounds are weighted about 1.5 dB higher. The LFE channel is excluded entirely, which surprises people the first time they see a 5.1 measurement that ignores the loudest-feeling part of the mix.
Why gating exists
If you averaged that over an entire programme, silence would drag the result down. A drama with long quiet passages would measure quieter than a continuous piece of music that actually sounds the same, and the fix would be to compress everything again.
So the measurement is gated. The programme is chopped into 400 ms blocks overlapping by 75 %, each block gets its own loudness value, and two thresholds are applied. Anything below −70 LUFS absolute is discarded outright, which removes true silence. Then the mean of what survives is computed, and everything more than 10 LU below that is discarded too. What remains is averaged, and that is the integrated loudness.
The second, relative gate is the important one. It is what makes the number describe the parts of the programme somebody is actually listening to, rather than the average of the loud parts and the room tone between them.
Integrated, short-term, momentary
You will see three windows named:
- Momentary — 400 ms, ungated. Useful live, on a meter, while mixing.
- Short-term — 3 seconds, ungated. What loudness range is built from.
- Integrated — the whole programme, gated as above. This is the number a delivery spec means when it says “programme loudness”.
A file has exactly one integrated value. Momentary and short-term are curves, not numbers, and quoting a single figure for either is a category error.
Loudness range, and what it is good for
Loudness range, LRA, is defined in EBU Tech 3342. Take the short-term loudness across the programme, throw away everything more than 20 LU below the gated mean, and report the spread between the 10th and 95th percentile of what is left. The unit is LU, and one LU is one dB.
It describes how much the loudness moves. A news bulletin might land around 3 or 4 LU. A feature film can sit above 15. An advert that has been squeezed flat will come in under 2.
Some specifications cap it, usually to stop a mix that is technically at target but swings so widely that dialogue becomes unintelligible on a phone speaker. A few ask for a minimum instead, to catch material that has been crushed. Both are blunt instruments. LRA is a distribution statistic and it does not know whether the loud parts were explosions or a badly ridden fader.
What LRA is genuinely useful for is spotting a file that does not belong. If nine deliverables from the same series measure between 5 and 7 LU and one comes in at 14, something happened to that one.
True peak is not the peak you can see
This is the part that catches people who have been mixing for years.
A digital audio file is a series of samples. Sample peak is simply the largest absolute sample value in the file, and it is what most meters show. But the samples are not the signal. They are points on a waveform, and the waveform between them is reconstructed by a filter in whatever converts the file back to sound. That reconstructed waveform can go higher than any individual sample.
The effect is called an inter-sample peak, and it is not exotic. A signal that peaks at exactly −0.1 dBFS on the sample meter can reconstruct to +0.5 dB or more, which clips the analogue stage on the other side. You will not see it in your editor. The listener hears it.
It gets worse after encoding. Lossy codecs do not preserve sample values; they preserve an approximation of the waveform, and the approximation routinely overshoots. A file that was sitting comfortably at −0.3 dBFS can come out of an AAC encoder with peaks above zero. This is why loudness specifications almost always carry a true-peak ceiling as a separate requirement, typically −1 dBTP and sometimes −2 dBTP.
Measuring it properly means oversampling. BS.1770-4 describes a 4× oversampling filter, which reconstructs enough of the waveform between samples to find the overshoots. The result is reported in dBTP, and it can and often does exceed the sample peak. A measurement that reports the two as identical is not measuring true peak.
Reading a set of numbers
Suppose a file comes back as:
- Integrated: −23.1 LUFS
- Loudness range: 6.4 LU
- True peak: −1.6 dBTP
Against a typical European broadcast spec of −23 LUFS ±1 with a −1 dBTP ceiling, that passes on all three. The integrated value is a tenth of a unit from target, the range is unremarkable for spoken content, and the peak has half a decibel of headroom against the limit.
Now suppose the same file reports −23.1 LUFS and −0.2 dBTP. The loudness is fine. The peak is not, and no amount of arguing about how good it sounds will change that, because the requirement exists to protect a stage that is not in your room.
And a file at −18.4 LUFS with a −3 dBTP peak is the opposite problem: plenty of headroom, four and a half units too loud. Somebody mixed it to a streaming target and delivered it to a broadcaster.
Doing this over a folder rather than one file at a time
Measuring one file is not the problem. The problem is thirty files from four suppliers, where two of them were mixed to the wrong target and nobody finds out until playout.
FileLister measures programme loudness, loudness range and true peak to BS.1770-4 and puts all three in the catalogue as ordinary columns, so you can sort a folder by loudness and see the outliers immediately. Because it is a measurement over every sample rather than a header read, it runs on demand over a selection rather than during a scan.
The results are also usable as requirements in a filter set, which is where this stops being a reporting exercise. A rule that says true peak must not exceed −1 dBTP and integrated loudness must sit between −24 and −22 turns a folder into a pass-or-fail list, and the failing files can be exported with the reason attached.
What the numbers do not tell you
Compliance is a floor, not a verdict. A file can hit −23.0 LUFS exactly, sit at −1.5 dBTP, and still be a bad mix — dialogue buried, music fighting the voice, a hard limiter chewing the transients. Loudness measurement catches the class of problem where a file is objectively wrong for the platform it is going to. It does not have opinions about anything else, and it should not be asked to.