Loudness on streaming services is a moving target with a fixed answer. The fixed part: every major platform normalises playback towards a target, so a master pushed to -6 LUFS is simply turned down — and still sounds flattened. The moving part is the handful of decibels between Spotify, Apple Music and YouTube, plus the fact that their encoders, not your DAW, get the last word on your peaks.
What follows is the checklist worth running through before any upload: target loudness, true-peak headroom, and the listening checks that catch what meters miss.
What the platforms actually do with your master
Loudness normalisation turns loud tracks down and, in some cases and on some playback settings, turns quiet tracks up. It happens at playback rather than on upload, and it's switched on by default almost everywhere. The practical upshot is that you can't win a loudness war on streaming. You can only lose dynamics.
The targets, in broad strokes:
- Spotify normalises to roughly -14 LUFS integrated, with playback settings that let listeners pick a quieter or louder mode.
- Apple Music Sound Check aims nearer -16 LUFS integrated.
- YouTube sits around -14 LUFS integrated, with its own attempts at evening out volume between videos.
- Tidal, Amazon Music and Deezer fall in the same -14 to -16 region. None of them reward you for going hotter.
Aim for about -14 LUFS integrated with a true-peak ceiling of -1 dBTP and you'll land sensibly on all of them. Treat that as a compass bearing, not a destination.
The pre-master checklist
Work in this order and you'll sidestep most of the trouble.
- Finish the mix first. If you're reaching for a limiter to fix a muddy low-mid or a vocal that won't sit, go back and fix the mix. Mastering can't un-bake that cake.
- Leave room on the mix bus. Peaks around -6 dBFS give the mastering chain somewhere to work. A mix that's already clipping has already made the decision for you.
- Set your target in the meter before you touch the limiter. Set it afterwards and you'll mix to the number instead of your ears.
- Watch short-term and momentary loudness, not just integrated. Integrated tells you where the track lands; short-term tells you whether the chorus is working much harder than the verse.
- Keep the ceiling at -1 dBTP. Drop to -1.5 or -2 if the master is dense or heading for a lossy codec at a low bitrate.
- Check mono. Fold the master down and listen. Anything that vanishes or turns hollow will do the same on a phone speaker or a club rig.
- Export 24-bit WAV. No dithering yet, unless you're reducing bit depth — and then dither once, at the very end.
Dynamics are the thing you're actually selling
Loudness range (LRA) is a decent gut check. Most modern productions sit somewhere between roughly 5 and 10 LU. If yours measures 2 or 3, you've almost certainly over-limited: transients are flattened, kick and bass are fighting each other, and the mix turns fatiguing after a minute.
Try this. Bypass the limiter and set the master so it peaks at -1 dBTP with nothing else changed. Then bring the level back up gradually and stop at the point where the snare still has a front edge and the vocal still breathes. In practice that's often a couple of decibels below where you were heading.
The limiter is the last decision you make, not the first.
Clipping can buy back a little level, but use it sparingly, on a single element, and only if you genuinely like what it does to the top end.
True peaks, codecs and the distortion you hear later
Your DAW shows sample peaks. Streaming services encode to AAC, Opus or similar, and those codecs reconstruct a waveform between the samples. Peaks that didn't exist in your WAV can appear after encoding, and they clip.
That's why a true-peak meter matters more than a sample-peak meter, and why -1 dBTP is the usual recommendation rather than 0. To catch the rest:
- Export a lossy version at a similar bitrate to the platform's, then play it back in the app your listeners actually use.
- Listen at low volume on a phone speaker. Codec artefacts and harsh limiting both show up there first.
- Check the quietest section too. Normalisation lifts quiet passages, which can expose noise, room tone or a reverb tail that was previously hidden.
Listening checks before you upload
Meters tell you what happened; speakers tell you whether it worked.
- Phone speaker, earbuds, laptop speakers, car, plus one decent system you trust.
- The whole track, start to finish, at a normal listening level. No skipping.
- Mono, twice: once quietly, once loud.
- Your three reference tracks, level-matched, in the same session. If yours sounds noticeably brighter or thinner next to them, fix that rather than the number.
- The first ten seconds. Listeners decide quickly, and a hot master sounds worst right there.
Delivery details worth getting right
Upload 24-bit WAVs at the session's native sample rate — unnecessary sample-rate conversion only adds risk. Apple's own mastering requirements ask for 24-bit or better sources, so there's no reason to send anything less.
Don't normalise twice. If your distributor offers loudness normalisation and the file is already mastered, leave it alone. Keep filenames tidy with the track title and version, and check ISRCs, credits and metadata before you publish, because fixing them afterwards is a chore. Trim long lead-in silence as well: a drawn-out gap before the first note makes the opening feel broken on a phone.
Trust the room, not the number
Master to a target, then stop measuring and start listening. Play the finished file on the worst speaker you own, in the car, on the headphones you use for the commute. If the low end holds together, the vocal stays in front, and nothing pinches when the chorus arrives, you're done. The LUFS figure will be whatever it is — and on streaming, that's fine.
Photo: Torsten Dettlaff / Pexels
