Audio Loudness Targets by Platform

Loudness Targets and True-Peak Ceilings

Two different kinds of number are in that table, and confusing them is the expensive mistake. A playback target is applied to you: you cannot miss it, and fighting it only wastes effort. A delivery spec is a contract you must hit, where being 3 LU out is a rejected master. Netflix, EBU R 128, ATSC A/85 and Apple Podcasts are contracts. Everything else on the list is something done to your file after you have handed it over.

Note also which rows are marked as unpublished. TikTok has no specification at all, Instagram deliberately has no single figure because Meta manages loudness against the listening context rather than a constant, and Amazon, Tidal and YouTube publish nothing even though their behaviour is measurable and stable. Any chart that lists those five with the same authority as EBU R 128 is telling you something untrue about how much you can rely on them.

The -19 LUFS figure widely quoted for mono podcasts is not Apple's, despite where you usually see it. It comes from Google's audio loudness guidelines, and it exists because a mono file duplicated to two speakers picks up 3.01 LU that the single-channel measurement never saw. The offset is real; the attribution usually is not.

LUFS, dBFS, RMS and True Peak: Four Units, Four Questions

dBFS answers "what is the largest number in this file". It is a sample measurement against digital full scale, and it correlates with perceived loudness so weakly as to be useless for the purpose: a sine wave and a snare hit can peak identically and sound nothing alike. Peak-normalizing two videos to -1 dBFS can leave them 15 LUFS apart, because one is a dense music bed and the other is a quiet voice with a single door slam in it.

RMS answers "how much raw power is in this window". It is an average rather than a single sample, which is a real improvement, but it treats 40 Hz and 3 kHz as equally loud. They are not. A bass-heavy mix and a speech mix at the same RMS will not sound remotely equal.

LUFS answers "how loud does this sound to a person". It is defined by ITU-R BS.1770 and used by EBU R 128, and it fixes RMS in two ways. The signal is K-weighted first — a high shelf of roughly 4 dB above about 2 kHz for the acoustic effect of a head in a sound field, and a high-pass that rolls off the bottom, where energy contributes far less to loudness than its raw power suggests. Then the measurement is gated, so quiet blocks are discarded rather than averaged in. LKFS, the unit in the American broadcast standard, is the same number under a different name; 1 LU equals 1 dB in both.

True peak (dBTP) answers a question none of the others do: "how high does the waveform go between the samples". Digital audio stores points, not a curve; the converter in a listener's phone reconstructs the curve, and the highest point of that curve is normally not one of your samples. The gap runs 1 to 3 dB. This is why every ceiling on the chart is -1 or -2 rather than 0, and why the resulting distortion is completely inaudible on your own timeline and obvious on somebody else's device: it is generated after your file, inside their decoder.

Integrated, Short-Term, Momentary

Every LUFS number needs a window attached to it or it means nothing. Momentary loudness is measured over 400 milliseconds. Short-term is measured over 3 seconds. Both are ungated and both exist for watching a meter while you work. Integrated is the whole programme with the gates applied, and it is the only one any platform normalizes against.

The gating is what makes the integrated number match human judgement. An absolute gate throws away every 400 ms block below -70 LUFS, which removes digital silence and room tone. The survivors are averaged, and a relative gate then throws away everything more than 10 LU below that average. What remains is the foreground of the programme. Without the relative gate, a talking-head video with long pauses would measure several LU quieter than it plainly sounds — and the obvious fix would be to compress the speech, which is exactly the wrong response. It also means a video whose integrated loudness is -14 LUFS can contain a short-term passage at -6, and that is not a fault. When a spec says -14 LUFS, it means integrated unless it says otherwise.

What Loudness Normalization Actually Does When You Upload

It measures the integrated loudness of your file, compares it to the platform's target, and applies one gain offset to the whole thing at playback. That is the entire operation. Your file is not re-encoded louder, no compression is added, and the dynamic range you delivered is preserved exactly.

The direction matters more than the number. YouTube, Apple Music, Amazon Music and Tidal only turn content down. Spotify moves in both directions. So there are two asymmetric outcomes, and only one of them is a real problem. If you export at -9 LUFS by squashing everything, YouTube applies -5 dB and you land at the same -14 as the person who did nothing — except your transients are gone permanently and theirs are not. If you export at -23 LUFS, nothing is applied at all, and you play back 9 dB quieter than the next video in the feed. On phone speakers in a room with other people in it, that difference is the difference between audible and not.

One more thing normalization does not do: it does not level your video internally. It is a single offset applied to the whole file, so if your intro is 10 LU louder than your outro, the platform preserves that gap perfectly. Internal consistency is a mixing problem — ducking, per-layer gain, riding the voice — and no loudness target solves it.

The Loudness War, and Where It Went

For roughly two decades, the way to sound better than the record next to you was to be louder than it. Because radio, CD players and early streaming all played back whatever level was in the file, compression and limiting bought a real competitive advantage, and masters climbed from around -18 LUFS in the eighties to -6 and beyond by the late 2000s. The cost was paid in transients: every dB of extra average level came out of the difference between the loud parts and the quiet ones.

Loudness normalization ended that arithmetic. Once the platform sets the final playback level, being louder in the file cannot make you louder in the listener's ear — it can only make you louder before the offset that removes exactly that advantage. What survives normalization is not your level. It is your dynamic range and your headroom.

The war has not ended everywhere, because the platforms that only turn content down leave a real incentive to sit at or slightly above the target rather than below it. But the incentive to go far past it is gone, and every dB past it is now a straight loss. This is the practical reason to measure before you master: the correction is usually a single gain offset, and the number tells you which direction it goes.

Why -14 LUFS Is a Sensible Default for Spoken-Word Web Video

There is nothing acoustically special about -14. It is roughly where a well-made modern master lands when it has not been compressed into a wall, which means most content clears it without the platform intervening. Spotify adopting it is where the number entered common use, and the others converged because a listener switching between apps hears a mismatched target as a bug. The convergence is what makes it useful, not the value.

For a talking-head video specifically, three things follow. First, it is the target for the platform most of that content is watched on, and the one platform that will not rescue you if you undershoot. Second, the gap to every other social and streaming target is at most 2 LU, which is inside what most listeners will notice at all — so one master genuinely serves all of them and per-platform exports are wasted work. Third, speech has a naturally wide loudness range and -14 leaves room for it; pushing to -11 or -9 to feel "competitive" means compressing a voice that was already loud enough, on a platform that is about to undo the level anyway.

The exceptions are worth knowing. If your primary distribution is podcast apps, master to -16 LKFS: Apple asks you to hit it, and Spotify will apply +2 dB on playback with no harm done. If you are delivering to a broadcaster or a streaming service under contract, use their spec and a dedicated meter, because -23 LUFS and -27 LKFS are not preferences. And do not chase the last tenth of a LU in any direction — every spec on this page has a tolerance wider than the difference is audible.

How to Hit a Loudness Target

1. 1
   Finish the mix first

   Integrated loudness is a whole-programme measurement, so it is only valid once nothing else will change. Adding a music bed after mastering invalidates the number, and appending a silent tail drags it down.
2. 2
   Measure two numbers, not one

   You need the integrated LUFS of the programme and its highest true peak in dBTP. Any EBU R 128 meter reports both, as does ffmpeg's loudnorm filter in analysis mode: ffmpeg -i input.mp4 -af loudnorm=print_format=json -f null -.
3. 3
   Apply the difference as one gain offset

   Target minus measured is the offset. Apply it to the whole file. It is a linear operation — it changes level and nothing else. The calculator above does this arithmetic and shows the resulting true peak.
4. 4
   Only limit if the ceiling is actually breached

   Check the true peak after the offset. If it now exceeds -1 dBTP (or -2 for Netflix, Amazon and ATSC A/85), a true-peak limiter brings it back. This is the only stage that alters the sound, so the less it has to do, the better.

Being within about 1 LU of the target is on-spec everywhere on this page. Chasing the last few tenths is not audible and is not worth a second render.

Mastering Inside the Edit

Valmera is an agentic AI video editor: you upload real footage, describe the edit in plain English, and an agent performs it. Loudness is one of the things you can ask for. Saying "master it" normalizes the finished mix to -14 LUFS integrated with a -1.5 dBTP ceiling, applied identically to the preview and the export, so what you approve is what ships. The ceiling is half a dB tighter than most platforms ask for, on the reasoning that they are going to re-encode the file to a lossy codec and that overshoot has to land somewhere.

It is one target and one toggle — no field for -16 or -23, and no LUFS readout in the interface — which covers social and streaming delivery and does not cover broadcast or film. It also changes loudness and not balance: if music is burying your voice, the fix is ducking the music under speech, not raising the master. The mechanics are in the music and audio docs, and the measurement itself is explained in full in the LUFS and loudness normalization glossary entry.

Frequently Asked Questions

Master Your Export Without Opening a Meter

50 free credits on signup, no card. Upload your footage, say "master it to -14 LUFS", and check the preview.