Why peak level was never enough
Peak level says how close the loudest single instant comes to the ceiling. It says nothing about the other four minutes. Two tracks can both peak at exactly 0 dBFS and differ by 12 dB in how loud they feel, because one of them spends most of its time near that ceiling and the other only touches it in the chorus.
That gap is what drove the loudness war: since peak was the only published limit, everything else was fair game, and the way to sound louder than the next record was to compress until the whole track sat at the top. LUFS closed it by measuring the thing listeners actually notice — and once the platforms started turning tracks down to a common loudness, the incentive to win that race went with it. That is the single change most mastering advice on the internet has not caught up with, and the practical guide works through what it means at the limiter.
The other older number, RMS, was closer to the mark: it averages the signal’s power over time, so it does respond to density rather than to single peaks. What it lacks is any idea of hearing. A low rumble and a vocal at the same RMS are nowhere near the same loudness to a person, and RMS cannot tell them apart. LUFS is, put simply, RMS with ears.
How the measurement works
The measurement is defined in ITU-R BS.1770, and it does three things. First it filters the audio to approximate how the head and the ear respond: a gentle shelf that lifts the upper frequencies by about 4 dB and a high-pass that rolls off the lowest bass, together called K-weighting. That is why a 60 Hz rumble does not count for as much as a vocal at the same level, and why a bright master reads louder than a dark one at the same RMS — if two masters measure the same and one feels louder, comparing their spectra usually explains it in a few seconds.
Second, it measures the mean power of the filtered signal in 400-millisecond blocks that overlap by three quarters, so nothing falls between two windows. In a multichannel programme the channels are summed with fixed weights — the surround channels count a little more, and the LFE channel is left out entirely.
Third, it throws away the quiet blocks before averaging. That step is called gating, it is the part people are most surprised by, and it gets its own section below.
Momentary, short-term and integrated
A loudness meter shows three numbers built from the same blocks, and they answer different questions:
- Momentary (M) is the loudness of the last 400 milliseconds. It moves constantly, and it is the one to watch when you want to know what a specific hit or phrase is doing.
- Short-term (S) averages the last three seconds. Slow enough to read, fast enough to follow a section — this is the reading a mastering engineer glances at through a chorus.
- Integrated (I) is one number for the whole programme, from start to finish, with the gating applied. It is the figure a streaming service measures and the only one of the three that a target like “−14 LUFS” refers to.
Most confusion about LUFS comes from mixing these up — reading a momentary peak of −8 on a track whose integrated loudness is −13, and concluding it is five decibels over target. It is not. The integrated reading is the whole story as far as normalisation is concerned.
LUFS vs dB, and what LU means
A decibel is a ratio, not a level — it only means something relative to a reference. dBFS uses digital full scale as that reference, so 0 dBFS is the loudest sample a file can hold and everything else is negative. LUFS uses the same scale and the same reference: the “FS” is the same full scale. The difference is what is being measured against it — a K-weighted, time-averaged loudness rather than an instantaneous sample value.
Because both use decibels, a change of 1 LU is exactly a change of 1 dB. LU is simply the unit for a difference in loudness, the way you would say a track is “3 LU over target” rather than “3 LUFS over”. LKFS, which you will meet in broadcast documents from the United States, is the same unit under an older name.
So there is no conversion between LUFS and dB, because there is nothing to convert. A track at −14 LUFS integrated might peak at −1 dBFS or at −6 dBFS depending on how dense it is; the two numbers describe different things about the same file.
Gating: why silence does not count
Two gates run before the blocks are averaged. The absolute gate discards anything below −70 LUFS — genuine silence, the count-in, the tail after the last chord. The relative gate then takes the average of what is left, and discards every block more than 10 LU below it.
The relative gate is what makes the number useful. Without it, a track with a long quiet intro would measure quieter than the same track trimmed, which would be a measurement of the editing rather than the music, and a spoken introduction before a song would drag the song’s own loudness down. With it, the integrated figure describes the loud parts the listener actually hears as “the track”.
It also explains a common surprise: fading the intro down or adding a quiet bridge does almost nothing to the integrated reading. If the loud sections are the same, the number is nearly the same, however much quiet material surrounds them.
Loudness range
Loudness range, LRA, comes from a companion document (EBU Tech 3342) and describes how much the short-term loudness moves across a programme, in LU. It takes the distribution of three-second readings, gates it, and reports the spread between the tenth and the ninety-fifth percentile — so a single explosive hit or one silent bar does not stretch it.
A heavily limited pop master might read 3 or 4 LU; an orchestral recording might read 15 or more. Neither figure is wrong. LRA describes the music, and the only real use for it in mastering is as a sanity check — if a record that was dynamic in the mix reads 2 LU after mastering, something was taken that probably should not have been.
True peak, and why your DAW disagrees
A digital file stores samples, not a waveform. The waveform is what comes back out of the converter, and it passes through the samples rather than stopping at them — so between two samples just under the ceiling, the reconstructed signal can be above it. Your DAW’s peak meter usually reports the samples; the encoder has to reproduce the curve.
This is why a master that showed −0.1 dBFS can distort audibly after upload. Measuring true peak means reconstructing between the samples — the standard oversamples by four — and reporting the highest point of the curve, in dBTP. It is the one number on a loudness meter that is a genuine warning rather than an observation, and it is why every platform target below comes with a ceiling around −1 dBTP: the lossy encode needs that margin. Distortion already printed into the file is a different question again, and the clipping detector is the one that answers it.
LUFS for Spotify, Apple Music and YouTube
Each major service measures your track’s integrated loudness on ingest and adjusts playback toward its own target. The published figures, and where each one comes from:
- Spotify — -14 LUFS integrated, ceiling -1 dBTP. Spotify Loudness Normalization documentation.
- Apple Music — -16 LUFS integrated, ceiling -1 dBTP. Apple Digital Masters technical specification.
- YouTube — -14 LUFS integrated, ceiling -1 dBTP. YouTube loudness normalisation, as measured and widely reported.
- Broadcast (EBU R128) — -23 LUFS integrated, ceiling -1 dBTP. EBU R128.
Two things these numbers are not. They are not upload limits — a track at −8 LUFS is accepted, and simply played 6 dB quieter than it was delivered on a platform targeting −14. And they are not symmetrical: Spotify raises a quiet track as far as its headroom allows, while YouTube and, as far as Apple documents it, Sound Check only turn tracks down. Nor are they a promise about the listener: normalisation is a setting on most services, some listeners switch it off, and a track played back raw is heard at whatever loudness it was mastered to. The streaming loudness targets guide goes through what each platform does on either side of its target, and there are pages for the two that get asked about most: LUFS for Spotify, including what the quiet, normal and loud settings each do, and LUFS for Apple Music, where the wrinkle is that Sound Check is a setting rather than a default.
What LUFS should you master to?
The honest answer is that there is no single number, and the people who give you one are usually answering a different question. If the platforms normalise anyway, mastering to −14 gains you nothing over mastering to −10 on a normalised stream — both arrive at −14. What changes between the two is everything else: the dynamics kept or spent, how the transients feel, whether the record breathes.
So the material decides. A sparse acoustic record wants to keep its range and will sit well under any streaming target; a dense electronic record can be mastered louder without losing anything it had, and will be turned down without being harmed. What every master should respect, regardless of genre, is the true-peak ceiling — because that is the one limit that causes audible damage rather than a volume adjustment.
If you want a reference rather than a rule, measure two or three released records in the same genre that you think sound right, and note where they land. Where they land is a fact about that genre’s taste; the target published by a platform is a fact about playback. Only one of them should decide how you master. Mastering hip-hop works one genre through in detail, including why the number that deserves strictness there is true peak rather than loudness.
How to measure it
Any meter that implements BS.1770 will give the same integrated figure to within a tenth of a LU, so the choice of tool matters less than reading the right number. Most DAWs ship one; the loudness checker on this site runs the same measurement in your browser, without uploading the file, and shows integrated loudness, true peak and loudness range against the platform figures above.
Measure the finished file, not the session: dither, the final limiter setting and any sample-rate conversion all belong in what is measured — the delivery step is where those get decided. And measure the whole track: a reading taken over the chorus is a short-term figure, whatever the meter labels it.
What to do if you are over target
Usually nothing. Being 3 LU over Spotify’s −14 means Spotify plays your track 3 dB quieter than it would otherwise — that is the whole consequence. Turning it down yourself achieves the same playback level while giving up whatever dynamics you spent to get loud in the first place.
The case that does need action is true peak above the platform’s ceiling, because that distorts in their encode after normalisation, where you will never hear it until somebody else does. If the master is over the ceiling, the fix is at the limiter: a true-peak-aware ceiling a decibel or so under full scale, and a small amount less gain into it.
If you would rather have the target set for you, that is what mastering does — Euphona’s AI mastering analyses the mix, chooses a loudness and true-peak target for the material rather than for a platform, and shows every move it made so you can read the reasoning back rather than take it on trust.
Keep reading