DSP
Foundational6 min read

How to read a loudness report: LUFS, true peak, and LRA

The three numbers that describe how loud a record actually is, what they measure, why louder is not automatically better — and what they said about a real album before and after mastering.

Learned building WPAudio Engine

Written and engineered by

Every mastering conversation eventually produces three numbers: an integrated loudness, a true peak, and a loudness range. They look interchangeable with “volume” and they are not. This article explains what each one measures, where the definitions come from, and then reads them against a real record — the twelve tracks of Wings & Prayers, before and after mastering.

A note on evidence, because this site is strict about it: everything in the first half of this article is general technical knowledge, and each claim carries its source. Everything in the second half is our production data, measured from the exact audio files published on this site. The two are labelled so you never have to guess which is which.

Integrated loudness: LUFS

Loudness is a perceptual quantity — how loud something sounds — and ITU-R BS.1770 defines the standard algorithm for estimating it from a digital signal. The signal is frequency-weighted (the “K-weighting” curve, which approximates how the ear weights spectral energy), mean-square energy is measured per channel, channels are summed, and quiet passages are excluded by a gate so that long silences cannot drag the average down. The result over a whole programme is integrated loudness, reported in LUFS — loudness units relative to full scale, where one loudness unit corresponds to one decibel.

Two properties make it useful. It matches perception far better than sample peaks do, and it is defined precisely enough to be reproducible: two compliant meters given the same file agree to within the tolerances EBU Tech 3341 sets for its test signals. That reproducibility is what lets a loudness value function as an engineering claim rather than an opinion.

Broadcast built its entire levelling practice on this measurement: EBU R 128 normalises programmes to −23 LUFS, and the accompanying EBU Tech 3341 defines the momentary, short-term and integrated meter modes. Music mastering is not broadcast and targets differ by context and intent — but the measurement is the same wherever you meet it.

True peak: dBTP

A digital file stores samples; a listener hears the reconstructed analogue waveform between them. The reconstruction can swing higher than any stored sample — an inter-sample peak — which means a file whose samples never touch full scale can still clip a converter or a lossy encoder downstream.

True-peak metering estimates the reconstructed peak by oversampling the signal (the algorithm in Annex 2 of ITU-R BS.1770, specified at 4× oversampling for 48 kHz material) and reports it in dBTP — decibels relative to full scale, true peak. This is why a mastering ceiling is stated in dBTP rather than in sample peaks, and why EBU R 128 permits a maximum of −1 dBTP: it leaves margin for what happens to the signal after you stop controlling it.

Loudness range: LRA

Integrated loudness says nothing about contrast. A record that whispers and then roars can share an integrated value with a record that holds one level throughout. Loudness range, defined in EBU Tech 3342, captures that difference: it is the spread between the 10th and 95th percentiles of the short-term loudness distribution, after gating, reported in LU. Low LRA means the programme sits at a consistent level; high LRA means it moves.

For mastering, LRA is the number that answers the accusation implicit in every loudness increase: did you flatten it? If loudness rises while LRA holds, the dynamics of the performance survived. If LRA collapses, they did not.

Why louder is not better

Play two versions of the same material and the louder one tends to be judged “better” — punchier, clearer, more exciting — even when the only difference is gain. That bias, and the decades of escalating release loudness it fuelled, is documented in Earl Vickers’ “The Loudness War: Background, Speculation, and Recommendations” (AES 129th Convention, 2010). The practical consequences run through the AES loudness project’s resources, and loudness normalisation on playback platforms has removed much of the prize: a hot master turned down to a platform’s reference level keeps its casualties — lost transients, reduced contrast — and gains nothing for them.

The honest way to compare two masters is therefore level-matched: attenuate the louder version by the measured loudness difference, then switch. Whatever preference survives level matching is about the mastering. Whatever preference does not survive it was about the gain. This is exactly why the A/B instrument on this site has a level-match toggle, and why it states the attenuation it applies.

The numbers on a real record

From here on, everything is our production evidence. The measurements below were taken from the exact listening files this site serves — the original pre-masters and the v1 reference masters of Wings & Prayers — using ffmpeg’s ebur128 filter, which implements the BS.1770 family of measurements. They describe this album. They are not universal targets, and they are not advice.

Reading the pre-masters, the three numbers told three different stories:

  • Integrated loudness exposed the album’s two populations. The core eight tracks sat within a 0.8 LU window (−16.1 to −15.3 LUFS) — mixed as one record. The four remixes spanned 4.6 LU (−18.2 to −13.6 LUFS) — arriving with no shared level at all.
  • True peak exposed the headroom situation: 3.5 to 5.1 dB of margin across the core album, but as little as 1.4 dB on the remixes.
  • The hardest track on the record was visible in two numbers before anyone pressed play: track 09 measured quietest (−18.2 LUFS) with only 1.6 dB of peak headroom — the longest distance to travel, and the least room to do it.

After mastering, the same three numbers described what the engineer did: every reference master peaks at −1.2 or −1.1 dBTP (one ceiling for the whole record), the album’s loudness spread narrowed from 4.6 to 2.7 LU, and no track’s loudness range moved by more than 0.7 LU. Louder, coherent, and not flattened — and each of those three words is a measurement, not a review. The full per-track tables are in the mastering case study.

Reproduce it

Nothing above requires our tooling. Given any audio file and ffmpeg:

ffmpeg -i track.wav -filter_complex ebur128=peak=true -f null -

The summary block reports integrated loudness (I), loudness range (LRA), and true peak — the same three numbers, from the same definitions, on your own material. A measurement you cannot reproduce is an anecdote; this one you can.

References

Keep reading