Someone sends back a mix and says it is four decibels too quiet. Someone else says the same file is three too loud. Both are reading a real number off a real meter. They are comparing it to different targets, and neither of them said which.
Loudness targets are not a single industry standard. They fall into two groups that work in opposite ways, and most of the confusion comes from treating them as one.
Broadcast: a target you must hit
Broadcast loudness is compliance. A programme is delivered at a specified integrated loudness, within a stated tolerance, and a file outside that window can be rejected before anyone listens to it.
The two families you will meet:
- EBU R 128 — European. Target −23 LUFS. Tolerance is normally ±0.5 LU for material that can be measured over its whole length, relaxed to ±1 LU for short-form or live. True peak ceiling −1 dBTP.
- ATSC A/85 — North American. Target −24 LKFS. LKFS and LUFS are the same measurement under two names; the unit was renamed and the older label stuck in the American document. Tolerance is typically ±2.
One decibel separates the two targets, which sounds trivial and is not, because the tolerances are tight enough that delivering a European mix to an American broadcaster puts you at the edge of the window before anything else goes wrong.
Individual broadcasters layer their own shims on top. Loudness range caps, dialogue-gated variants, tighter peak ceilings at −2 dBTP. The spec you were sent governs, not the standard it is based on.
Streaming: a target that gets applied to you
Streaming platforms do not reject a file for being loud. They turn it down.
Every major music and video service measures what you upload and applies a gain offset at playback so that everything in a listener's queue sits at roughly the same level. The published figures — the ones people quote as targets — are the level the platform normalises to, not a requirement you must meet.
The numbers cluster loosely around −14 LUFS, with variation between services and, on some of them, between playback contexts. They also change. Treating any specific figure as permanent is how people end up mixing to a value a platform abandoned two years earlier.
What actually matters is the consequence of being on the wrong side of it:
- Louder than the platform's target and it will be turned down. Everything you gained by compressing harder is removed, and what remains is a mix with less dynamic range than the one next to it, played at the same level. This is the mechanism that ended the loudness war, and it ended it by making the war pointless rather than by banning it.
- Quieter than the target and most services will leave it alone rather than turn it up, because raising level risks clipping. So a quiet master stays quiet next to everything else.
The practical position: aim somewhere near the platform's figure, keep true peak below −1 dBTP so the normalisation gain does not push anything into clipping, and stop optimising past that.
Why the same file cannot serve both
A broadcast master at −23 LUFS uploaded to a streaming service is nine decibels below where that service normalises. Some platforms will leave it quiet. A streaming master at −14 delivered to a European broadcaster is nine decibels over target and will be bounced.
This is not a mixing problem so much as a versioning one. Programmes that go to both usually need two deliverables, and the mistake is almost never made deliberately — it is made when the wrong file gets picked out of a folder that contains both.
Which is why loudness belongs in the inventory of a delivery folder rather than in someone's memory. If the catalogue of a folder shows integrated loudness beside each file, a mix at −14 sitting among mixes at −23 is visible in a glance rather than at playout.
Cinema, and why it uses a different unit entirely
Theatrical mixing is referenced to a calibrated room, not to a loudness target in the file. The reference is a fader position — 7.0 on a Dolby processor — and a pink noise calibration to 85 dB SPL per channel. There is no LUFS figure to hit, because the playback gain is fixed and known.
This matters when theatrical material is repurposed. A cinema mix has enormous dynamic range and no loudness normalisation, and dropping it into a broadcast or streaming chain without a new pass produces exactly what you would expect: dialogue nobody can hear, followed by something that makes the neighbours knock.
The true peak ceiling is a separate requirement
It is worth stating on its own, because it is the one people treat as a formality.
The ceiling is not there to control loudness. It is there because the file will be encoded again, probably more than once, and lossy encoding overshoots. A file sitting at −0.2 dBTP going into an AAC encode can come out clipping. A −1 dBTP requirement gives the encoder somewhere to put the overshoot.
This is also why sample peak is not an acceptable substitute for the measurement. Sample peak reads the values in the file. True peak reconstructs the waveform between them at 4× oversampling and reports what a converter would actually produce. On dense material the two can differ by more than a decibel, always in the direction that gets you rejected.
Reading a spec you have been sent
A usable loudness clause names four things. If one is missing, ask rather than assume:
- The target, in LUFS or LKFS. “Broadcast standard” is not a target.
- The tolerance. ±0.5 and ±2 are very different jobs.
- The true peak ceiling, in dBTP. If it is quoted in dBFS, the spec was written before 2010 or by somebody copying one that was.
- Whether loudness range is constrained, and to what.
Two further things are worth confirming for anything long-form. Whether the measurement covers the whole file or a defined programme segment, since bars and tone at the head will move an integrated figure. And whether the gating is standard or dialogue-gated, which some broadcasters specify and which produces a different number on the same file.
Checking a delivery rather than trusting it
Most loudness failures are not mixing errors. They are the wrong file, the wrong version, or a mix that was correct for the platform it was originally made for. All three are caught by measuring what is in the folder before it leaves, and none of them are caught by asking whether anyone remembered.
FileLister measures integrated loudness, loudness range and true peak across a whole folder and keeps them as columns beside the codec, frame rate and duration. A filter set can then encode the spec directly — integrated between −24 and −22, true peak at or below −1 — and the files that miss it come out as a list with the reason beside each one.
The value is not the measurement. It is finding out on your own machine rather than from the person you delivered to.