Both services change the level of your master at playback, and they do it in opposite directions. Spotify normalizes to -14 dB LUFS and applies gain both ways, so a quiet master gets lifted. Apple's Sound Check only attenuates, so the same quiet master stays quiet next to everything around it. Deliver one file at -14 LUFS integrated with true peak under -1 dBTP, or under -2 dBTP if the master runs hotter than -14, and both platforms behave predictably. Below is where those numbers come from, what the encoder does to them, and which widely quoted figure is missing from the vendor's own documentation.
What does Spotify actually do to the file?
Nothing to the file. Spotify's loudness documentation is explicit that loudness is measured at upload and gain is applied during playback, to the ITU 1770 standard, with the target at -14 dB LUFS. An album is normalized as one unit so the quiet track stays quiet against the loud one. Individual tracks are normalized when they turn up in shuffle or a playlist, which is where most catalogue listening happens.
Premium listeners get three settings, and the numbers are published: Loud at -11 LUFS, Normal at -14, Quiet at -19. On Loud, Spotify inserts a limiter that engages at -1 dB in sample values with a 5 ms attack and a 100 ms decay. That limiter is not yours and you cannot hear it while you work, which is one more reason to leave the last dB alone.
The rule that catches people is on the way up. Positive gain is capped by the headroom in the file. Spotify's own example: a track measuring -20 LUFS with a true peak of -5 dBFS is lifted only to -16 LUFS, not to -14, because 1 dB of headroom is reserved for the lossy encode. A quiet master with one stray peak near zero gets stranded 2 dB below the catalogue it is sitting in.
Where does the -16 LUFS figure for Apple Music come from?
Not from Apple. The published Apple Digital Masters document describes a delivery chain rather than a loudness target: a 24-bit source at its native rate, encoded to 256 kbps AAC, verified with two free Apple tools. afclip reports which channel clipped, when, and by how much. AURoundTripAAC lets you audition the encoded result against the source before anyone else hears it.
The -16 figure is the community's measurement of Sound Check behaviour, repeated until it looks like a spec. Treat it as a description, and treat the direction as the part that matters: Sound Check turns loud material down and does not raise quiet material. A master sitting at -18 LUFS stays 4 dB under a -14 LUFS catalogue on Apple Music and gets lifted to roughly -17 on Spotify. Same file, two different outcomes, and only one of them can be fixed by making the master louder.
Why does true peak matter more than sample peak?
Because the waveform the listener hears is reconstructed after decoding, and it does not pass through the same points as the file. Ogg Vorbis and AAC both produce inter-sample peaks that sit above the highest sample value in the source. A master reading 0.0 dBFS on a sample-peak meter can decode above 0 dBFS and clip inside a phone's DAC, where you have no access to it.
Spotify asks for below -1 dBTP, and below -2 dBTP if the master is louder than -14 LUFS, on the grounds that louder tracks are more susceptible to distortion in transcoding. Measure with a true-peak meter running at least 4x oversampling, which is the minimum in ITU-R BS.1770, then encode and listen. A round-trip check through AAC at 256 kbps takes under a minute per track and catches the one chorus that reads clean on the meter and buzzes on the decode.
Is -14 LUFS a standard, or a house number?
A house number. ITU-R BS.1770 defines how loudness is measured, using K-weighting and a two-stage gate, and says nothing about what to aim for. The engineering body that did make a recommendation reached a different answer: AES TD1004.1.15-10 proposed a target window of -16 to -20 LUFS for internet streaming with a maximum of -1 dBTP, and it was superseded by AES TD1008 in September 2021. Platform targets sit in the neighbourhood of that window without matching it.
The practical consequence is that there is no single correct number to master to, only a range where playback gain does the least damage. Anywhere between -16 and -13 LUFS integrated with peaks controlled will survive every current normalizer. Below -18 you start losing on Apple. Above -11 you are handing several dB of your own limiting to a system that would have turned you down anyway.
What improves when you stop mastering at -8?
Crest factor, and it is measurable. A master at -8 LUFS turned down 6 dB by the platform arrives at -14 LUFS with the transient damage already printed into it. A master built at -14 arrives at the same loudness with the drum hits intact. Watch short-term LUFS rather than integrated while you work, because integrated hides a chorus that is 3 dB over the verse.
One concrete threshold to work to: if the limiter shows more than 3 dB of gain reduction on the loudest chorus, the fix belongs in the mix. Usually it is a kick and a bass overlapping between 60 and 100 Hz, or a vocal bus compressor that has already flattened the dynamic range before the master chain sees it. Pull 2 dB out of the mix bus and the limiter stops working for a living.
What do you deliver, and how many versions?
One file per track: 24-bit WAV at the session sample rate, no rate conversion, no separate platform masters. Every service normalizes the same measurement, so a second file cut louder for one of them does nothing except create a version-control problem six months later. Note the integrated LUFS and the true peak value in the delivery sheet so the distributor and the label are reading the same numbers you are. Full credit list with sources at /credits.
Keep one alternative with 2 to 3 dB less limiting for sync work. An editor cutting music under dialogue will pull the level anyway and wants the headroom, and a heavily limited master fights the ducking curve rather than sitting under it. Label that file clearly in the folder name, because the one thing worse than sending the wrong master is not knowing which one went out.
Frequently asked
Should I master to -14 LUFS for Spotify? Aim for -14 LUFS integrated with true peak below -1 dBTP, which is what Spotify's own guidance asks for. Anything between -16 and -13 LUFS with controlled peaks plays back at the same level, so treat -14 as the centre of a range rather than a target to hit exactly.
Does Spotify turn quiet tracks up? Yes, and Apple Music does not. Spotify applies positive gain up to -14 LUFS but caps the lift so 1 dB of headroom survives the lossy encode, so a -20 LUFS track peaking at -5 dBFS only reaches -16 LUFS. Apple's Sound Check attenuates loud material and leaves quiet material where it is.
Why is -1 dBTP recommended instead of -0.1 dBFS? Because lossy encoding produces inter-sample peaks above the highest sample in the file. A master reading 0.0 dBFS on a sample-peak meter can decode above zero and clip in the playback device. Measure with a true-peak meter at 4x oversampling or higher, per ITU-R BS.1770.
If a master is going out this week
Measure the integrated LUFS and the true peak, run one round trip through 256 kbps AAC, and listen to the loudest chorus on the decode rather than the source. If the limiter is pulling more than 3 dB there, send the mix back before the master goes to the distributor. Background and contact at /about.