Loudness, measured rather than guessed · Editing Video

Why peak level is the wrong measurement

Two files can both peak at -1 dBFS and one can be twice as loud as the other. Peak level measures the single highest sample. Perceived loudness depends on how much energy there is over time, and on which frequencies it sits in, because the ear is far more sensitive around 2–5 kHz than at the extremes.

So a meter that only shows peaks tells you whether you will clip. It tells you nothing about whether your video is louder or quieter than the one before it in the feed.

LUFS

LUFS — Loudness Units relative to Full Scale — is a measurement that applies a filter modelling the ear's frequency response and then averages energy over time. It was standardised as ITU-R BS.1770 and adopted in Europe as EBU R128. It is the number every platform now normalises against.

Three readings, and they answer different questions:

Integrated — the average over the whole programme. This is the headline number, and it is what platforms use.
Short-term — a three-second window. Useful for spotting a section that is too quiet.
Momentary — a 400 ms window. Useful for finding individual loud moments.

There is also loudness range (LRA), which measures how much the loudness varies across the piece. A very low LRA means everything is at the same level, which is fatiguing. A very high one means quiet sections will be inaudible on a phone.

True peak, separately

dBTP — true peak — estimates the peak of the reconstructed analogue waveform rather than the highest digital sample. These differ: between two samples the actual waveform can rise above both, a phenomenon called inter-sample peaking.

A file that peaks at exactly 0 dBFS can therefore clip when it is converted to analogue, or when it is re-encoded to a lossy format, which every platform does. Leave headroom: -1 dBTP is the normal target.

The numbers platforms use

Roughly, as of writing, and these move:

YouTube: normalises toward around -14 LUFS.
Spotify, Apple Music: around -14 to -16 LUFS.
Instagram, TikTok and short-form: broadly similar territory.
European broadcast (EBU R128): -23 LUFS, with true peak at -1 dBTP. This is a regulatory standard, not a suggestion, and broadcasters reject files outside tolerance.
US broadcast (ATSC A/85): -24 LKFS, which is the same measurement under a different name.

The gap between -14 and -23 is large, and it is the reason a file that passed broadcast QC sounds quiet on YouTube. Ask where the file is going before you set the level.

What normalisation means for you

This is the part that changes behaviour.

Platforms turn loud uploads down. They generally do not turn quiet ones up by much, if at all. So there is no advantage to mastering loud, and there is a real cost: to get a mix to -6 LUFS you have compressed and limited it heavily, destroying its dynamic range. The platform then attenuates it by 8 dB, and you arrive at the same perceived volume as everyone else with a flatter, more tiring mix.

Mix to roughly the target and let the platform leave you alone.

Measuring it, free

Resolve's Fairlight page has a loudness meter with integrated, short-term, momentary, LRA and true peak, and a preset for each broadcast standard. Open it, play the whole programme from the start, and read the integrated value at the end. It must be the whole programme — an integrated reading of thirty seconds is a reading of thirty seconds.

FFmpeg measures it without opening anything:

bash

ffmpeg -i film.mp4 -af ebur128=peak=true -f null -

The summary at the end gives integrated loudness, range and true peak. And FFmpeg will also correct it, in two passes for accuracy:

bash

ffmpeg -i film.mp4 -af loudnorm=I=-14:TP=-1:LRA=11:print_format=summary \
       -f null -
# then feed the measured_* values back into a second pass

The two-pass form matters: a single-pass loudnorm works dynamically and can audibly alter the mix, while two-pass applies a single fixed gain, which is what you want on a finished programme.

YouLean Loudness Meter has a free version that runs as a plug-in in any DAW or NLE.

Dialogue-anchored loudness

One refinement worth knowing. Broadcast practice anchors loudness to the dialogue rather than to the whole mix, because a film with loud music and quiet speech can measure correctly and still be unintelligible.

The practical version: get your dialogue sitting comfortably first, then build music and effects around it, then check the integrated figure. If you have to pull the whole mix down to hit the target, pull the music down rather than the dialogue.

Today

Run the ebur128 command above on the last video you published. Compare its integrated figure to -14. Whatever the number is, it explains how your video sounded next to everyone else's.