Loudness, LUFS and True Peak

For decades, "loud" in mastering meant "as loud as possible". Streaming changed that. Most services now normalise playback: they measure each track's loudness and turn louder tracks down so listeners don't reach for the volume between songs. To master well today you need to understand how that measurement works.

Peak level vs loudness

A peak meter shows the highest sample value, in dBFS. It tells you whether you're about to clip, but almost nothing about how loud a song sounds: a single snare hit can peak at −1 dBFS in a quiet ballad.

Loudness is measured in LUFS (Loudness Units relative to Full Scale), defined in the ITU-R BS.1770 standard. It weights the signal to roughly match human hearing — less sensitive to deep bass, more to the upper mids — and averages it over time. One LU equals one dB, so "two LU louder" means what you'd expect.

The three time windows

Loudness meters show several readings because they average over different lengths:

• Momentary (M): a 400 ms window. Jumps around with every hit; useful for spotting a sudden jump in level.
• Short-term (S): a 3-second window. Good for comparing sections — how much louder is the chorus than the verse?
• Integrated (I): the whole song, start to finish, with very quiet passages gated out so silence and fades don't drag the number down. This is the number streaming services use to set playback level.

There's also loudness range (LRA), which describes how much the short-term loudness varies across the song. A heavily compressed EDM track might have an LRA of 3–5 LU; a film score can be over 15.

True peak

Digital audio is a series of samples, but the analogue waveform rebuilt from them can rise between the samples. These inter-sample peaks don't show on a sample-peak meter, yet they can clip a digital-to-analogue converter or — more commonly — a lossy encoder.

A true-peak meter (dBTP) oversamples the signal to estimate those in-between peaks. Lossy encoding (MP3, AAC, Ogg Vorbis) tends to raise peaks further, so a master that touches 0.0 dBFS sample peak can clip once a streaming service encodes it.

Practical rule: keep true peak at or below −1 dBTP. For very loud masters, −2 dBTP gives the encoder more margin — Spotify's own guidance asks for −2 dBTP on masters louder than −14 LUFS.

Streaming reference levels

Each platform sets its own playback reference, but not every platform publishes it. These are the figures at the time of writing — the services change them, so check before a release:

Platform	Playback reference (integrated)	Where the figure comes from
Spotify	−14 LUFS (Normal setting)	Spotify's own documentation
YouTube	about −14 LUFS	Independent measurement; YouTube doesn't publish a figure
Apple Music (Sound Check)	about −16 LUFS	Independent measurement; Apple doesn't publish a figure

How normalisation behaves differs too. A track mastered at −8 LUFS on Spotify is simply played about 6 dB quieter. A track quieter than the reference may be turned up only as far as its peaks allow, or not at all, depending on the service and the listener's settings.

What this means for how loud to master

There is no rule that says you must master to −14 LUFS. What normalisation changes is the reward for extra loudness:

• Louder than the reference costs you dynamics — punch, transients, contrast between sections — and the service turns you down anyway. You end up at the same playback level as everyone else, but flatter.
• Much quieter than the reference can leave your song sounding weaker next to its neighbours if the service doesn't turn it up.

Many engineers therefore master to somewhere between the reference and what the genre expects: a dense club track might land around −8 to −10 LUFS integrated because limiting is part of its sound, while an acoustic record might sit at −14 to −16 and keep its dynamics. Pick the loudness that serves the music, check that true peak is under −1 dBTP, and don't chase a number for its own sake.

Reading a meter while you work

1. Play the loudest section and watch short-term loudness — that's what the limiter is working hardest on.
2. Play the whole song once to get integrated loudness.
3. Check the maximum true peak after the limiter.
4. Note the loudness range. If it's collapsed to 2–3 LU on a song that should breathe, you've over-limited.

SoundGen's Master view shows live loudness while the song plays, so you can watch these numbers move as you change the chain.