Mastering

What loudness should you master to?

Master to about −14 LUFS integrated with true peaks no higher than −1 dBTP, and you are done. That is what Spotify publishes and normalises to, in both directions. Here is the part almost every other article leaves out: Spotify and Apple Podcasts are the only two services that publish a loudness number at all. The tidy table of targets for Apple Music, YouTube, Amazon and Tidal that you have read a hundred times is third-party measurement, not a specification any of those companies issued.

By MELD · 12 September 2026 · 8 min read

What is LUFS, and why not just use decibels?

A decibel meter tells you how big the waveform is. It does not tell you how loud the track sounds, because human hearing is not equally sensitive at every frequency, and because a single peak says nothing about the two minutes around it.

LUFS measures perceived loudness over time. The measurement is defined by ITU-R BS.1770, currently at revision 5, in force since November 2023 — worth knowing, because most mastering articles still cite BS.1770-4, which has been superseded for nearly three years. In Europe the same measurement is wrapped in EBU R 128, version 5.0, also November 2023.

One naming quirk that trips people up: ITU calls the unit LKFS and EBU calls it LUFS. They are the same unit. Apple writes LKFS, Spotify writes LUFS, and a figure in one is directly comparable to a figure in the other.

The number that matters for a release is the integrated loudness — one figure for the whole track. That is what a streaming service reads, and what it uses to decide how much to turn you up or down.

What does Spotify actually do to your track?

Spotify is the one service that documents this properly, so it is worth quoting exactly. From its loudness normalization page:

• Louder than −14 LUFS: negative gain is applied, bringing the track down to −14.
• Quieter than −14 LUFS: positive gain is applied, bringing the track up to −14. Normalisation runs both ways, which surprises people who assume a quiet master is simply left alone.
• Albums are normalised as a whole, so the relative loudness between your tracks survives. Individual tracks are adjusted when someone shuffles, or plays across several albums.

Listeners on Premium can override the target: Loud is −11 LUFS, Normal is −14, Quiet is −19. On the Loud setting Spotify applies its own limiter — threshold −1 dB on sample values, 5 ms attack, 100 ms decay — because there is no headroom left to turn things up into.

The practical consequence: if you master at −8 LUFS to sound competitive, Spotify turns you down 6 dB and the listener hears a squashed track at the same volume as everyone else. You spent your dynamic range and got nothing back.

Which services publish a target, and which do not?

This is the table you actually want, with the column every other version omits.

The last four rows are measurements people have taken, repeated until they hardened into specification. They are probably in the right area. They are not something those companies committed to, which means they can change without an announcement, and it means an article presenting them as official is wrong about what it is telling you.

There is one more wrinkle worth knowing, because it explains why sources disagree. The Audio Engineering Society's technical document AESTD1008 — which superseded the older TD1004 back in 2021, though most articles still cite TD1004 — recommends −16 LUFS for track-normalised music and −14 for album-normalised. So the industry body and the biggest streaming service do not agree with each other. Broadcast is further away still: EBU R 128 targets −23 LUFS.

Two more from Apple, one from Sound Check, one specific to spoken word. Apple Music's Sound Check is a setting the listener turns on, not an always-on server-side process like Spotify's. Apple Podcasts, by contrast, publishes a hard delivery spec.

So what number should you actually use?

−14 LUFS integrated, true peak at or below −1 dBTP. That single setting is inside the tolerance of every target above, published or measured, and it is what Spotify itself asks for.

The −1 dBTP ceiling is the one figure everybody genuinely agrees on. It comes from AESTD1008: "For all content, it is recommended that the Maximum True Peak level not exceed −1 dBTP at the codec input of lossy-encoded streams." EBU R 128 says the same, and so do Spotify and Apple Podcasts independently.

The reason is encoding. MP3 and AAC are lossy: the decoded waveform is not the one you handed over, and it can overshoot. Leaving a decibel of room means the overshoot lands below full scale instead of clipping in a listener's player, where you cannot hear it and cannot fix it.

If you deliberately master louder than −14, Spotify's own advice tightens: keep true peaks below −2 dB, because "very loud masters with true peaks above −2 dB can introduce distortion during encoding".

Does mastering louder still help anywhere?

On Spotify, no. The gain is applied on the way out and the loudness war ends in a draw at −14 LUFS for everyone.

For the services that publish nothing, the honest answer is that nobody outside those companies knows for certain whether they turn quiet tracks up as well as loud ones down. The widely repeated claim that YouTube only ever attenuates is plausible and consistent with what people measure, but it is not in any document Google has published. Treat it as folklore that is probably true rather than a rule you can rely on.

One real, and rarely mentioned, way to lose loudness for nothing: Spotify notes that "inaudible high-frequencies in your mix can cause the loudness algorithm to overestimate the perceived loudness, because ITU 1770 doesn't apply a lowpass filter". Ultrasonic rubbish you cannot hear can make the meter read louder than the track sounds, and the service then turns you down for it.

What about podcasts and spoken word?

Different number, and this one is published. Apple Podcasts asks for −16 dB LKFS ±1 dB, the same for mono and stereo, measured to ITU-R BS.1770-5, with true peak not exceeding −1 dBFS.

Speech and music genuinely need different targets. AESTD1008 puts the reason plainly: "speech normalized to the same BS.1770 Integrated Loudness as music is typically perceived 2 to 3 dB louder than the music." Deliver a talk programme at the music target and it will feel shouty next to everything else.

How do you measure this without buying a meter?

You need two things: a true-peak-aware LUFS reading of what you have, and a way to hear the before and after at the same loudness. The second matters more than people expect. Anything louder sounds better for about four seconds, so comparing a master to its original at different volumes tells you nothing except which one is louder.

MELD Master does both in a browser tab. It measures the file, renders to the target you pick, shows five checks — loudness, true peak, dynamic range, clipping, DC offset — each passing or failing with a reason, and plays the original against the mastered version at matched loudness so you are judging the work rather than the volume. Nothing is uploaded: the measuring, the mastering and the download all happen on your own machine. The first render is free and needs no account.

One thing mastering cannot do: un-clip. If the file you start with is already clipped, a limiter cannot put the tops of the waveform back. That is why the clipping check runs on the source as well as the result.

Working backwards from here: the mix you are mastering has to come from somewhere, and recording it take over take in a browser is a shorter distance than it used to be.