Loudness for broadcast and streaming — LUFS, true peak, targets
Loudness is no longer measured by peaks but by how loud a programme sounds. The targets differ according to where the video is going — and if you don’t meet them, the platform will turn you down itself.
From peaks to loudness
Sound for television used to be normalised by its peaks: nothing was allowed to exceed a certain level. You know the result — adverts squeezed to the maximum by a compressor were much louder than the film, even though the peaks were the same. In R 128 the EBU states plainly that loudness differences between programmes and channels are the most common reason for viewer complaints.
The solution is to measure integrated loudness — an average of how loud a programme sounds to the ear — using the ITU-R BS.1770 algorithm. The unit is LUFS (Loudness Units relative to Full Scale); the ITU and ATSC use the label LKFS, which is the same quantity. A difference of 1 LU = 1 dB.
What a loudness meter measures
| Quantity | What it means | What it is for |
|---|---|---|
| Integrated (I) | The loudness of the whole programme, with gating (silence isn’t counted) | The value compared with the target, e.g. −23 LUFS |
| Short-term (S) | A sliding 3 s window | Continuous monitoring while mixing |
| Momentary (M) | A sliding 400 ms window | Fast changes, checking maxima |
| Loudness Range (LRA) | The difference between the quieter and louder parts in LU | The dynamics of a programme; not recommended for programmes shorter than 1 min |
| True Peak | The highest value of the signal waveform, including inter-sample peaks, in dBTP | Protection against distortion in D/A conversion and lossy compression |
Targets by delivery
DIAGRAM · TARGET INTEGRATED LOUDNESS
European broadcasting — EBU R 128. Programme loudness is normalised to −23.0 LUFS. Where the target can’t practically be reached (in live programmes, for example), a tolerance of ±1.0 LU is permitted; in quality control ±0.2 LU is allowed for measurement errors. The maximum true peak is −1 dBTP in production; it may be lower for some distribution systems. The whole programme is measured, not just the dialogue. Short-form content (adverts, promos) is covered by the R 128 s1 supplement.
USA — ATSC A/85. For exchanging content without metadata it recommends a target of −24 LKFS. The recommendation is the basis for complying with the US CALM Act. On 8 July 2026 the ATSC published the first major revision since 2013 (A/85:2026-07), which adds an annex for streaming services and a short quick-reference guide to the values.
Netflix (original productions). According to specification v1.6: near-field mixes in both 5.1 and 2.0 at −27 LKFS ±2 LU, measured on dialogue (dialogue-gated) according to ITU-R BS.1770 across the whole programme, true peak no higher than −2 dBTP, 48 kHz / 24-bit, with no lossy compression. Theatrical mixes have no LKFS requirement.
Broadcaster streaming — EBU R 128 s2. The EBU recommends streaming programmes unchanged at −23 LUFS and using loudness metadata. If a broadcaster wants to control the dynamics itself and doesn’t use metadata, it may temporarily raise the distribution loudness to −20 to −16 LUFS.
YouTube and other platforms. In its help pages YouTube says that it may adjust loudness levels automatically, and it offers a Stable volume feature that evens out quiet and loud passages. However, YouTube gives no specific LUFS target in its official documentation. The values circulating on forums are third-party measurements — so we don’t present them here as a specification.
Platform normalisation usually turns a loud mix down, but doesn’t necessarily turn a quiet one up (to avoid creating peaks). So an over-compressed, ‘loud’ mix won’t be louder after normalisation, just flatter. A quiet mix will sound quiet.
How to prepare the mix
- Intelligible dialogue first. Dialogue is the anchor — viewers set their volume by the speech. If they have to turn down for music and effects and turn up for dialogue, the mix doesn’t work, whatever the number says.
- Mix at a fixed monitoring level. Calibrated monitoring (Netflix gives a reference level of 79 or 82 dB SPL for near-field) is more reliable than watching the meter.
- Measure the whole programme. The meter measures integrated loudness over the full length; make corrections by changing the overall gain, not with a last-minute compressor.
- A true-peak-detecting limiter on the output, with the ceiling set for the target (−1 dBTP for R 128, −2 dBTP for Netflix).
- Fit the dynamics to the medium. Cinema and home cinema can take a large LRA; a phone on a tram can’t. If you make several versions, make them deliberately.