Published: 2026-06-05 · Reading time: 21 min read · Author: Nikolay Sapunov, CEO at Fora Soft

Why this matters

If you build or operate a service that delivers sound — a streaming platform, an OTT app, a video conferencing product, a podcast host, a broadcaster's playout chain — someone on your team is being handed a delivery spec that says "−23 LUFS, −1 dBTP, measured per BS.1770" and is expected to know what that means. This article is for the product manager, founder, or operations lead who needs to understand those numbers well enough to write them into a contract, reject a non-compliant deliverable, or explain to a regulator why your ads are not louder than your programmes. By the end you will be able to read any loudness spec in the industry, know which of the three standards governs your market, and run the loudness measurement arithmetic yourself. Every number here traces to the controlling document — the ITU-R Recommendation, the EBU R-document, or the ATSC Recommended Practice — not to a vendor's blog.

The problem loudness normalization solves

Start with the experience everyone has had. You are watching a quiet film, the dialogue is calm, and then an advertisement arrives that feels twice as loud even though the volume knob has not moved. You lunge for the remote. That jump is the problem the entire field of loudness normalization exists to kill.

The reason the ad sounds louder is that, for decades, audio was leveled by its peaks — the single highest instantaneous sample in the signal. A peak meter answers the question "did the signal clip?" and nothing else. Two tracks can have the exact same peak level and yet one sounds far louder, because loudness is not about the highest moment; it is about the average energy the ear is exposed to over time. An advertiser who compresses a track — squashing the quiet parts up toward the loud parts — keeps the same peak but raises the average enormously. The peak meter says both tracks are fine. The ear says one is shouting.

Loudness normalization replaces "level by peak" with "level by perceived loudness". Instead of asking "what was the loudest instant?", it asks "how loud did this sound on average to a person?", produces one number for the whole programme, and nudges every programme to the same number. Once every piece of content is normalized to the same loudness, the volume knob can stay put. That single idea — measure perceived loudness, normalize everything to one target — is what the three standards in this article implement.

If LUFS, peak, RMS, and the basic vocabulary of loudness are new to you, read the foundations piece first: Loudness, peak, RMS, LUFS — measuring how loud audio actually is. This article assumes you know what a LUFS number is and focuses on the standards that define and target it.

The three documents and how they fit together

People talk about "the loudness standard" as if it were one thing. It is three things, and they do different jobs. Confusing them is the single most common mistake in delivery emails, so it is worth fixing the picture before any numbers.

The first document, ITU-R BS.1770, is the measurement algorithm. ITU-R stands for the Radiocommunication Sector of the International Telecommunication Union, the global body that standardizes how broadcast signals are measured. BS.1770 is a recipe: feed it a soundtrack, and it returns a single number describing the programme's loudness on a scale called LUFS — Loudness Units relative to Full Scale. BS.1770 does not tell you what number to aim for. It is the ruler, not the rule.

The second document, EBU R128, is the European target rulebook. EBU stands for the European Broadcasting Union, the alliance of European public broadcasters. R128 takes the BS.1770 ruler and says: measure with it, aim for −23 LUFS, and never let the true peak exceed −1 dBTP. It is the rule that uses the ruler.

The third document, ATSC A/85, is the United States target rulebook. ATSC is the Advanced Television Systems Committee, the body that defines American digital-television standards. A/85 also uses the BS.1770 ruler, but it aims for −24 LKFS and handles dialogue differently. It carries extra weight because a US federal law — the CALM Act — makes A/85 compliance mandatory for television commercials, and the Federal Communications Commission can fine a broadcaster for breaking it.

So the relationship is simple: one ruler (BS.1770), two regional rules built on top of it (EBU R128 in Europe, ATSC A/85 in the US). Hold that picture; the rest of the article fills it in.

Layered diagram showing ITU-R BS.1770 as the shared measurement ruler at the base, with EBU R128 at minus 23 LUFS and ATSC A/85 at minus 24 LKFS as two regional target rulebooks built on top of it Figure 1. One ruler, two rules. BS.1770 measures; EBU R128 and ATSC A/85 each set a target that uses that measurement.

ITU-R BS.1770: the ruler everyone measures with

BS.1770 — its full title is "Algorithms to measure audio programme loudness and true-peak audio level" — is the foundation. Both regional rulebooks defer to it, so understanding it once explains the measurement half of every loudness spec you will ever read. The current revision is ITU-R BS.1770-5 (November 2023). The algorithm has four steps, and you can follow each one without any audio engineering background.

Step 1 — K-weighting: hear the way an ear hears

The first step corrects for a fact about human hearing: we do not hear all frequencies equally loudly. A burst of energy at 3 kHz — roughly the pitch of a baby's cry or a smoke alarm — sounds much louder to us than the same energy down at 50 Hz, a low bass rumble. A microphone does not know this; it treats every frequency the same. So the algorithm first runs the signal through a fixed pair of filters, together called K-weighting, that boost the frequencies the ear is sensitive to and trim the ones it is not. After K-weighting, the numbers the algorithm sees line up with what a person would actually perceive as loud. This is why the LUFS scale carries a "K" in its broadcast spelling, LKFS — Loudness, K-weighted, relative to Full Scale. LUFS and LKFS are the same number with two names; Europe says LUFS, the US says LKFS.

Step 2 — mean square: turn a waveform into energy

A waveform swings positive and negative thousands of times a second; if you simply averaged it, the positives and negatives would cancel to roughly zero. So the algorithm squares every sample first — squaring makes every value positive and, conveniently, corresponds to the signal's power, the rate at which it delivers energy. It then averages those squared values. The result is the mean-square energy of the K-weighted signal: a single positive number that says how much perceptual energy the track carried.

Step 3 — gating: ignore the silence

Here is the step that separates BS.1770 from a naive average, and the one most people get wrong. Imagine a two-hour film with long silent passages — a held breath before a line, a quiet night scene. If you averaged those silences into the loudness, a film with a lot of silence would read as artificially quiet, and you would turn it up until the loud scenes blasted. Gating prevents this. The algorithm chops the programme into overlapping windows — 400-millisecond blocks, each overlapping the next by 75% — and throws away the windows that are too quiet to count, before averaging the rest.

It throws them away in two passes. The absolute gate discards any block quieter than −70 LUFS — that is effectively silence, and it never counts. Then the algorithm measures the average of what survives, and applies a relative gate: it discards any block more than 10 LU below that first average. (LU, Loudness Units, is the unit you use when talking about a difference on the LUFS scale; LUFS is the absolute value, LU is the gap between two values.) What remains — the blocks that actually carry the programme's body — is averaged one last time to produce the integrated loudness, the single number for the whole programme. Gating was added in BS.1770-2 (2011); the original 2006 version had none, which is why a measurement from a pre-2011 meter can disagree with a modern one.

Step 4 — true peak: catch the peak between the samples

The fourth step is separate from loudness and answers a different question: will this signal distort on playback? An ordinary peak meter looks at the digital samples and reports the highest one. But the real analog waveform that a speaker reproduces passes through and sometimes above those sample points — the true peak can sit higher than any single sample. BS.1770 computes the true peak by oversampling the signal (reconstructing the waveform at four times the sample rate) and finding the real maximum, reported in dBTP — decibels relative to full scale, true peak. The true-peak filter coefficients were refined in BS.1770-3 (2012). This is why a spec says "−1 dBTP" and not "−1 dBFS": it wants the true inter-sample peak held a hair below the digital ceiling, so nothing clips when the signal becomes analog again.

Four-stage signal-flow pipeline showing a waveform entering K-weighting filters, then mean-square energy, then two-pass gating that discards quiet blocks, producing an integrated LUFS value, with a separate true-peak oversampling branch producing a dBTP value Figure 2. The BS.1770 measurement pipeline: K-weight, square, gate, integrate — plus a separate true-peak branch. The output is one LUFS number and one dBTP number.

Working the arithmetic of a gate

The gating step is abstract, so here is the relative-gate calculation with real numbers. Suppose the absolute-gated average of a programme — every block above −70 LUFS, averaged — comes out at −21 LUFS. The relative gate sits 10 LU below that:

relative gate threshold = absolute-gated loudness − 10 LU
relative gate threshold = −21 LUFS − 10 LU
relative gate threshold = −31 LUFS

Now the algorithm re-averages, this time keeping only the blocks louder than −31 LUFS and discarding everything quieter. Say that final average lands at −20.4 LUFS. That −20.4 LUFS is the programme's integrated loudness — the number you compare against the target. Because the quiet−scene blocks were dropped, the figure reflects the loudness of the programme's actual body, not its silences.

EBU R128: the European rulebook

EBU R128 — "Loudness normalisation and permitted maximum level of audio signals" — is the document that turned loudness normalization from theory into European broadcast law-by-convention starting in 2010. It is short and prescriptive. It makes four demands, and they are worth knowing exactly.

First, the target is −23 LUFS, measured with BS.1770. Every normalized programme aims for this single number. Second, a tolerance of ±0.5 LU is allowed for content that can be processed precisely (most file-based delivery), widening to ±1 LU for less predictable material such as a live mix where you cannot rehearse the loudness. Third, the true peak must never exceed −1 dBTP. Fourth, R128 introduces a companion measure, the Loudness Range (LRA), which describes how much the loudness varies within a programme — a single quiet-to-loud spread figure, expressed in LU, so a delivery spec can ask for dynamics that are neither flat nor wildly uneven. LRA is defined in EBU Tech 3342.

R128 does not redefine the measurement; it points at BS.1770 and adds the practical scaffolding broadcasters need. That scaffolding lives in a family of companion EBU Tech documents: Tech 3341 defines the "EBU Mode" meter with its three readouts — Momentary (a 400-ms window for catching instantaneous spikes), Short-term (a 3-second window for following the mix as you work), and Integrated (the gated whole-programme number that must hit −23). Tech 3343 covers production practice, and Tech 3344 covers distribution, including how to hand loudness metadata to modern immersive codecs like AC-4 and MPEG-H.

The standard has kept pace with how people actually consume audio. The base recommendation is at version 5.0 (November 2023), and it now carries four supplements for contexts the original 2010 broadcast document did not foresee: R128 s1 for short-form content like adverts and promos, R128 s2 ("Loudness in Streaming", published November 2023) for on-demand and streaming distribution, R128 s3 for radio, and R128 s4 for cinematic content. The streaming supplement matters because phone-and-earbud listening tolerates — and often prefers — a louder target than a quiet living room, which is why streaming music platforms cluster nearer −14 LUFS than −23. We map every platform's actual number in LUFS targets per platform in 2026.

ATSC A/85 and the CALM Act: the American rulebook with teeth

The American story is the same algorithm with a different target and a legal enforcement mechanism that Europe's convention-based system lacks.

ATSC A/85 — "Techniques for Establishing and Maintaining Audio Loudness for Digital Television" — sets the US digital-television target at −24 LKFS, one unit quieter than Europe's −23. The current text is A/85:2013, with Corrigendum No. 1 (February 2021), and it measures with BS.1770-3. The one-LU difference from EBU R128 is historical and practical rather than perceptual; for any real workflow, −23 and −24 are close enough that a single mix delivered to both targets needs only a one-decibel trim.

A/85 differs from R128 in one substantive way: how it handles dialogue. R128 measures the whole programme. A/85 leans on the idea of the Anchor Element — the part of the soundtrack a listener instinctively uses to judge loudness, which in most television is the dialogue. A/85 normalizes to the loudness of that anchor, so a film with loud action scenes and quiet dialogue is leveled by its dialogue, the thing the viewer actually tracks. This anchor-based, dialogue-centric approach is also why American delivery specs talk about dialnorm (covered next) more than European ones do.

The reason A/85 carries more weight than a recommended practice usually would is the CALM Act — the Commercial Advertisement Loudness Mitigation Act, a US federal law in force since December 2012. The CALM Act does not write its own technical rules; it adopts ATSC A/85 by reference and makes compliance mandatory for television commercials. The Federal Communications Commission enforces it, and a broadcaster that lets ads run louder than the surrounding programme can be the subject of complaints and penalties. So in the US, loudness normalization is not a best practice a broadcaster may adopt — for commercials it is the law, and the technical definition of "too loud" is A/85's −24 LKFS.

Dialnorm: the metadata that carries the loudness number

One term shows up constantly in American and Dolby-based delivery and confuses everyone who meets it: dialnorm. It is short for dialogue normalization, and it is a small piece of metadata carried inside Dolby codecs (AC-3 and E-AC-3) that tells the playback device how loud the programme's dialogue is, as a positive number of decibels below full scale. If your programme's dialogue measures −24 LKFS and you deliver it untouched, you set dialnorm to 24. The decoder in the TV reads that 24 and automatically attenuates so that every programme's dialogue lands at the same comfortable level, regardless of how each was mixed.

The pitfall is brutal in its simplicity: if the dialnorm value is wrong — if it says 24 but the dialogue is actually at −31 — the decoder applies the wrong correction and the programme plays back at the wrong loudness, defeating the entire system. We cover the practical handling of dialnorm, replay gain, and encoder normalization in Loudness in practice: dialnorm, normalize, attenuate, replay gain.

Side by side: how the three standards compare

The standards agree far more than they differ, because they share the BS.1770 measurement engine. The table below is the whole comparison a delivery spec turns on.

Attribute ITU-R BS.1770 EBU R128 ATSC A/85
What it is Measurement algorithm European target rulebook US target rulebook
Current revision BS.1770-5 (Nov 2023) R128 v5.0 (Nov 2023) A/85:2013 + Corr. 1 (Feb 2021)
Loudness target none — it only measures −23 LUFS −24 LKFS
Unit name LKFS / LUFS LUFS LKFS
Measurement basis itself BS.1770 BS.1770-3
Dialogue handling gating only whole-programme Anchor Element (dialogue)
True-peak ceiling defines dBTP −1 dBTP −2 dBTP (recommended)
Tolerance n/a ±0.5 LU (±1 LU live) ±2 LU (typical practice)
Legal force none broadcaster convention mandatory for ads (CALM Act)

Table 1. The three loudness documents at a glance. The shared BS.1770 engine is why the targets sit one LU apart rather than in different worlds. Sources: ITU-R BS.1770-5; EBU R128 v5.0; ATSC A/85:2013.

Notice what the table makes obvious: the hard differences are the target number (−23 vs −24), the dialogue treatment (whole-programme vs anchor), and the legal status (convention vs law). Everything else is shared. A team that internalizes that can stop treating "EBU" and "ATSC" specs as alien and start treating them as the same measurement with two regional dials.

A worked normalization, end to end

Tie it together with a complete example. You receive a finished programme. Your meter, implementing BS.1770-5, reports an integrated loudness of −20.4 LUFS and a true peak of −0.3 dBTP. Your delivery spec is EBU R128: target −23 LUFS, ceiling −1 dBTP.

First, the loudness correction. You are 2.6 LU too loud, so you turn the whole programme down by that amount:

gain to apply = target − measured
gain to apply = −23 LUFS − (−20.4 LUFS)
gain to apply = −2.6 LU

You apply −2.6 dB of gain to the entire programme. Because you moved everything down by the same amount, the relationships inside the mix are untouched — only the overall level changed. Now re-check the peak. The true peak was −0.3 dBTP; lowering the level by 2.6 dB lowers the peak too:

new true peak = old true peak + gain
new true peak = −0.3 dBTP + (−2.6 dB)
new true peak = −2.9 dBTP

−2.9 dBTP is below the −1 dBTP ceiling, so the programme now passes both tests: −23 LUFS integrated, −2.9 dBTP true peak. Note the order matters. If the loudness correction had been upward — if the programme had been too quiet — raising the level could have pushed the true peak above −1 dBTP, and you would then need a true-peak limiter to pull the peaks back down without changing the integrated loudness. That ordering trap is the next pitfall.

Pitfall — normalizing loudness and peak in the wrong order, or confusing the two. Loudness (LUFS, an average) and true peak (dBTP, a maximum) are different measurements and you must satisfy both. The classic failure is to limit the peaks first and then normalize loudness upward, which re-introduces peaks above the ceiling; or to treat a peak number as if it were a loudness number. The rule: normalize integrated loudness to target first, then check the true peak, then apply a true-peak limiter only if the ceiling is exceeded — and re-measure loudness afterward, because heavy limiting can nudge it. We go deep on the inter-sample peak side of this in True peak, dBTP and the inter-sample peak problem.

A second pitfall: the −23 vs −14 confusion

The most damaging misconception in the field is "−23 LUFS is the loudness standard." It is the broadcast target. The number you are asked to hit depends entirely on where the content plays. European and US broadcast sits near −23/−24 LUFS because a living room with attentive listeners has low background noise and wide dynamics are welcome. Streaming music platforms normalize much louder — Spotify and others cluster around −14 LUFS — because phone-and-earbud listening in noisy environments favours a denser, louder, more consistent signal. Cinema is different again, calibrated by reference level rather than a LUFS target.

Deliver a podcast mastered at −23 LUFS to a platform expecting −16 and listeners will reach for the volume, exactly the behaviour normalization was supposed to end. The standards in this article define how to measure and give you the broadcast targets; the per-platform targets are a separate table, and getting them wrong is a common, avoidable error. The full 2026 set lives in LUFS targets per platform in 2026.

Where Fora Soft fits in

Across the OTT, video streaming, e-learning, telemedicine, and video conferencing products we have built since 2005, the loudness question surfaces the same way: a delivery spec or a partner integration arrives quoting "−23 LUFS, −1 dBTP, BS.1770", and the team needs the measurement built into the pipeline rather than checked by hand at the end. For OTT and streaming clients we wire a BS.1770 loudness measurement and a true-peak limiter into the transcoding step, so every asset is normalized to the spec's target before it reaches the packager. For e-learning and telemedicine, where dialogue intelligibility is the whole product, we normalize captured audio to a consistent target so a quiet lecturer and a loud one arrive at the listener at the same comfortable level. The standard does not change; only which target and which true-peak ceiling the project's market demands.

What to read next

Call to action

References

  1. ITU-R, Recommendation BS.1770-5, "Algorithms to measure audio programme loudness and true-peak audio level" (ITU-R, November 2023, accessed 2026-06-05). Controlling source for the four-step measurement algorithm: K-weighting, mean-square, the 400-ms / 75%-overlap gating blocks, the −70 LUFS absolute gate and −10 LU relative gate, and the true-peak (dBTP) oversampling computation. https://www.itu.int/rec/R-REC-BS.1770 — Tier 1 (official ITU-R Recommendation; the controlling measurement standard).
  2. EBU, R 128, "Loudness normalisation and permitted maximum level of audio signals", v5.0 (European Broadcasting Union, November 2023, accessed 2026-06-05). Controlling source for the −23 LUFS target, the ±0.5 LU / ±1 LU tolerances, the −1 dBTP true-peak ceiling, and the four supplements (s1 short-form, s2 streaming, s3 radio, s4 cinematic). https://tech.ebu.ch/publications/r128 — Tier 1 (official EBU Recommendation).
  3. EBU, Tech 3341, "Loudness Metering: 'EBU Mode' metering to supplement EBU R 128 loudness normalisation" (European Broadcasting Union, current revision, accessed 2026-06-05). Source for the Momentary (400 ms), Short-term (3 s), and Integrated (gated) meter definitions. https://tech.ebu.ch/publications/tech3341 — Tier 1 (official EBU Tech specification).
  4. EBU, Tech 3342, "Loudness Range: A measure to supplement EBU R 128 loudness normalisation" (European Broadcasting Union, current revision, accessed 2026-06-05). Source for the Loudness Range (LRA) definition in LU. https://tech.ebu.ch/publications/tech3342 — Tier 1 (official EBU Tech specification).
  5. ATSC, A/85:2013 (with Corrigendum No. 1), "Techniques for Establishing and Maintaining Audio Loudness for Digital Television" (Advanced Television Systems Committee, 2013; corrigendum February 2021; accessed 2026-06-05). Controlling source for the −24 LKFS target, the Anchor Element (dialogue) approach, the dialnorm definition, and the reference to BS.1770-3. https://www.atsc.org/atsc-documents/a85-techniques-for-establishing-and-maintaining-audio-loudness-for-digital-television/ — Tier 1 (official ATSC Recommended Practice).
  6. Commercial Advertisement Loudness Mitigation (CALM) Act and FCC implementing rules (US Federal Communications Commission, in force since December 2012, accessed 2026-06-05). Source for the legal status of A/85: the CALM Act adopts ATSC A/85 by reference and makes loudness compliance mandatory for television commercials. https://www.fcc.gov/enforcement/areas/sound-volume-commercials-calm-act — Tier 1 (US federal regulator; the controlling legal source).
  7. Netflix Studios Partner Help Center, "Loudness and True Peaks: How to Measure and When to Flag" (Netflix, accessed 2026-06-05). First-party deployer source for a real production delivery spec: −27 LKFS dialog-gated, ±2 LU, −2 dBTP, measured per BS.1770 — a worked example of how a platform applies the standard. https://partnerhelp.netflixstudios.com/hc/en-us/articles/360050414014 — Tier 4 (first-party engineering documentation of the deployer; cited as a deployment example, not as the source of the standard).
  8. EBU, R 128 s2, "Loudness in Streaming" (European Broadcasting Union, Geneva, November 2023, accessed 2026-06-05). Source for the streaming-distribution supplement to R128 and the rationale for louder targets in mobile / on-demand listening contexts. https://tech.ebu.ch/publications/r128 — Tier 1 (official EBU Recommendation supplement).
  9. T. Lund, "ITU-R BS.1770 Revisited" (TC Electronic / AES, accessed 2026-06-05). Peer-context source for the history of BS.1770 revisions, the introduction of gating in BS.1770-2, and the true-peak refinements in BS.1770-3. https://toneprints.com/media/1018237/lundt013011.pdf — Tier 5 (AES-presented technical paper; used only for revision history, never overriding the spec text).