Mastering for Streaming: LUFS Targets & File Specs

What does loudness normalization do to my master?

Streaming platforms don't re-encode your file to change its level. They analyze the master you deliver, store a loudness value as metadata, and apply a gain offset in the playback chain when someone hits play. Your audio file is untouched. The volume the listener hears is set at the moment of playback, on their device.

Most platforms only turn audio down. If your master is louder than the target, it gets attenuated to hit the target. Spotify is the main exception that also turns quiet tracks up in its Normal mode, and even then only within the headroom your true peak allows. Apple Music's Sound Check only ever reduces gain, so for Apple the target works as a ceiling, not a floor.

This is the fact that should drive every mastering decision for streaming. A heavily limited master pushed loud does not arrive loud. It arrives at the target like everything else, carrying whatever damage the limiting did to its transients. A more dynamic master played at the same normalized level can hit harder because its peaks were never crushed. You're mastering to sound good at the normalized level. That single shift is why the loudness war stopped making sense for streaming releases.

LUFS targets by platform

There's no single number, but they cluster tightly. Most platforms land on -14 LUFS integrated. Apple Music is the notable outlier at -16 LUFS, following the AES TD1008 album-level recommendation. A few sit a step off: Deezer normalizes around -15 LUFS, Amazon Music lands around -14 (sources vary between -13 and -14, and Amazon doesn't publish a clean spec). Spotify is the only one that lets the listener pick, offering Loud, Normal, and Quiet presets, with Normal at -14 as the default.

You don't need to deliver a different master per platform. Master once to a sensible target, keep your true peak in check, and let normalization handle the rest. The full per-platform breakdown, including the direction each platform applies gain and which use track versus album normalization, lives in the DSP loudness chart. Spotify's specific mechanics, including how its positive gain stops short when true peak headroom runs out, get their own detailed walkthrough in the Spotify LUFS target guide.

→

Why -1 dBTP is the ceiling everywhere

True peak and sample peak are two different measurements, and the gap between them is where a lot of clean-looking masters go wrong. A sample peak meter reads the level of individual digital samples. A true peak meter estimates the level of the reconstructed analog waveform between those samples, where it can overshoot the points it's drawn from. Those overshoots are inter-sample peaks, and they can run meaningfully higher than anything the sample meter shows.

The reason this matters for streaming: every platform delivers audio to listeners through lossy codecs like AAC and Ogg Vorbis. The encode-and-decode process introduces its own inter-sample behavior, so peaks can come out of the encoder higher than they went in. A master sitting right at the digital ceiling can clip after transcoding, even though it looked perfectly clean on your meters at home.

That's why EBU R 128, AES TD1008, and every major platform independently land on -1 dBTP. The dB of headroom costs you almost nothing in perceived loudness and removes a category of distortion you can't fix once the file is encoded. Setting it correctly means using a true peak limiter, not a standard sample-peak brickwall. Use a true peak limiter. The distinction matters, and the full explanation of inter-sample peaks and oversampling is in the true peak limits guide.

File specs: what distributors want

The safe delivery file is a 24-bit, 44.1 kHz stereo WAV. That works for every major distributor and matches the sample rate streaming platforms deliver at, so nothing gets converted in the chain. Higher sample rates like 96 or 192 kHz are accepted by some distributors and by Apple Digital Masters, but they buy the end listener nothing on standard streaming.

Two things trip people up. Don't upsample a 44.1 kHz master to a higher rate before delivery. An upsampled file is detectable and it's not the same as a native high-res recording. And don't dither a 24-bit master down to 16-bit yourself unless a distributor specifically requires it. Let the platform handle that conversion with its own dithering.

There are format quirks too. Spotify requires plain WAVE_FORMAT_PCM WAV rather than the extensible variant, and prefers FLAC for its lossless tier. Stereo is the standard. Mono files get flagged or rejected by some distributors unless the track is genuinely mono. Dolby Atmos is a separate immersive mix and ADM BWF delivery path with receiving-service ceilings, not another stereo loudness target.

→

The loudness war is over for streaming

For decades the incentive was to master louder than the competition, because on radio and CD the louder track simply sounded bigger at the same dial position. Normalization removed that incentive. A brickwalled -8 LUFS master and a dynamic -16 LUFS master both arrive at the listener at -14. The loud one gave up its dynamics for nothing.

The practical guideline follows from that: master to a level that sounds good once it's normalized. Plenty of engineers targeting streaming now work somewhere in the -11 to -14 LUFS range for pop, rock, and electronic, holding onto transient punch, with that -1 dBTP ceiling and without the sustained full-band limiting that defined the loud-war era. The exact target is a creative call within that range.