Guide
EBU R 128
What −23 LUFS and −1 dBTP actually require, read out of the recommendation itself rather than out of other people's summaries of it.
The numbers
| Measure | Requirement |
|---|---|
| Target loudness | −23.0 LUFS |
| Maximum True Peak Level | −1 dBTP |
| Tolerance on the target | ±1.0 LU, and only where the target is not practically achievable |
| Tolerance in quality control | ±0.2 LU, to account for measurement error |
| Tolerance on the true-peak measurement | ±0.3 dB |
| Measured with | ITU-R BS.1770, including the gating of equation (7), and EBU Tech 3341 |
From EBU R 128, version 5, November 2023. First published February 2010; drafted by the EBU's PLOUD group.
Why peak normalisation was abandoned
R 128 exists to replace the peak normalisation broadcasting used before it, and the recommendation states its own reasoning. Levelling by peak left considerable loudness differences between programmes and between channels, and those inconsistencies were the cause of most viewer and listener complaints. The quasi-peak programme meter it relied on does not reflect loudness at all, and was never designed to indicate a long-term average.
So R 128 is not a recommendation about sound quality. It is a recommendation about the moment one programme becomes another and the level does not change.
How −23.0 LUFS is measured
Programme Loudness is defined as the integrated loudness over the duration of a programme. Not a section of it. The recommendation is explicit that the audio shall generally be measured in its entirety, without emphasis on specific foreground elements such as speech, music or sound effects, so metering the dialogue and matching that is not what R 128 asks for.
The measurement is gated. The level gate defined in equation (7) of ITU-R BS.1770 measures foreground loudness, which the recommendation says is what makes programmes with a wide loudness range match each other better.
One thing worth knowing before comparing old figures: the relative gate was changed from −8 LU to −10 LU in August 2011, in the second version of the recommendation. Measurements from before that are not comparable with measurements after it.
−1 dBTP is a ceiling, not a safe place to sit
The requirement is that the True Peak Level of a programme shall not exceed −1 dBTP during production, for linear audio, measured with a meter compliant with ITU-R BS.1770 and EBU Tech 3341. Two qualifications sit in the same clause, and they are the parts usually left out.
The measurement carries a tolerance of ±0.3 dB for signals with a bandwidth limited to 20 kHz. Aiming at exactly −1.0 dBTP therefore means accepting that the meter's own error can put you over.
The permitted maximum can be lower elsewhere. The recommendation states that permitted Maximum True Peak Levels may be lower for different distribution systems and data reduction rates, and refers the reader to EBU Tech 3344. A file that measures −1.0 dBTP before encoding is not guaranteed to measure −1.0 dBTP after it: lossy coding reconstructs a waveform that can sit higher than the one that went in.
What a true peak is
The recommendation defines Maximum True Peak Level as the maximum value of the audio signal waveform of a programme in the continuous time domain.
That phrase is the whole of it. The measurement is not the largest sample value; it is the largest value the waveform reaches, including between the samples. This is why a peak meter reading below zero in a DAW can still correspond to a true peak above it, and why the two readings disagree more as the material gets louder and denser.
The three optional measures
Beside the target, R 128 says three further measures may be used. None of them is required.
Loudness Range (LRA) describes the distribution of loudness within a programme, measured per EBU Tech 3342. It is not recommended for programmes shorter than one minute: LRA is built from short-term values on a three-second window, and a short programme does not provide enough of them.
Maximum Momentary Loudness and Maximum Short-term Loudness, measured per EBU Tech 3341, can be used to decide whether a programme exceeds the upper loudness tolerance of its intended audience.
R 128 is the general case
The recommendation points at separate documents for particular uses, and the numbers there are not always the ones above.
| For | See |
|---|---|
| Short-form content (adverts, promos) | R 128 s1 |
| Streaming | R 128 s2 |
| Radio | R 128 s3 and EBU Tech 3401 |
| Cinematic content | R 128 s4 |
| Production | EBU Tech 3343 |
| Distribution and reproduction | EBU Tech 3344 |
Streaming is the one that catches people out. Guidance for streaming is in supplement 2, not in R 128 itself, so applying −23 LUFS to a streaming platform is reading the wrong document.
LUFS and LKFS
They are the same unit. ITU-R BS.1770 writes LKFS; the EBU writes LUFS, which it notes is the form compliant with international naming conventions. A meter labelled one way and a meter labelled the other are measuring the same thing.
Measuring it
What the recommendation requires of a meter is narrow: compliance with ITU-R BS.1770, including the gating of equation (7), and with EBU Tech 3341. Two meters that both satisfy that will agree on a given file.
Ours is Spectra, built on a BS.1770-5 engine, showing momentary, short-term and integrated loudness, loudness range and true peak at once. If all you want is a VU meter, SpectraVU is free and shares the same measurement core.
Source
Everything above is from EBU R 128, version 5, November 2023. Where this page and the recommendation disagree, the recommendation is right, please tell us and we will fix it.