Broadcast Loudness Normalization Without Tears: Getting BS.1770-4 Right Across a Multi-Station Deployment

By the KAVANA engineering team — June 2026

Loudness normalization is one of those broadcast topics where the theory is well-documented and the practice is genuinely hard. The standards exist — BS.1770-4, EBU R 128, ATSC A/85, and a collection of national variants — and they are reasonably clear about what they require. The problem is that implementing them consistently across a multi-station deployment, with heterogeneous source material, multiple content production pipelines, and varying monitoring infrastructure, requires engineering decisions that the standards do not make for you.

We have been implementing loudness normalization across our station customer base for the better part of a decade. This post is an honest account of what works, what does not, where the standards genuinely conflict with each other, and what we built to address the parts that existing commercial tools handle poorly.

Why the Standards Do Not Simply Solve the Problem

BS.1770-4 is the ITU standard for measuring audio loudness. It defines an integrated loudness measurement methodology — the LUFS (Loudness Units relative to Full Scale) metric — and a true peak measurement algorithm that accounts for inter-sample peaks that can cause distortion in digital-to-analog conversion. EBU R 128 builds on BS.1770 with specific target levels and tolerance windows: -23 LUFS integrated, ±1 LU tolerance, true peak maximum of -1 dBTP. ATSC A/85 is the North American variant targeting -24 LUFS.

The standards are coherent and technically sound. The difficulty is that they were developed primarily for television production and post-production environments, where content flows through a defined processing chain and every piece of audio can in principle be measured, adjusted, and verified before transmission. Broadcast radio, particularly in the format we support, does not work that way.

A county-level radio station in our customer base might be running: live presenter audio from a studio microphone through a hardware compressor whose gain staging has not been audited in three years; pre-produced news packages from an in-house production system; advertising spots from a dozen different agencies with wildly varying loudness profiles; automated AI-synthesized content from our pipeline; and networked content from a provincial feed that arrives as encoded audio with its own processing chain applied upstream. All of this is going to the transmitter. The BS.1770-4 target applies to everything.

Measuring and normalizing each of these sources to -23 LUFS is straightforward in isolation. Doing it consistently across all sources, in real time, for every station in a network simultaneously, without introducing audible artifacts, processing latency, or interactions between sources that cause level jumps at transitions — that is the engineering problem.

The Three Loudness Problems That Actually Matter

Before getting into implementation, it helps to distinguish three different loudness problems that are often conflated but require different solutions.

The legacy content problem. A significant fraction of the advertising spots in circulation at Chinese regional stations were produced before loudness normalization standards were widely adopted, or by agencies that did not apply normalization, or by clients who instructed their production houses to make the spots as loud as possible to stand out from surrounding programming. These spots can be 6 to 10 LUFS louder than the -23 target, and they vary significantly from each other. A listener who hears three consecutive spots that are -23 LUFS, -17 LUFS, and -20 LUFS experiences a loudness roller-coaster that degrades the listening experience regardless of whether each individual spot is technically compliant with its own production standard.

The live-versus-recorded problem. A hardware AGC on the studio microphone chain produces audio that is dynamically level-controlled but not loudness-normalized. If the presenter reads the same passage at different times of day with different energy, the LUFS integrated over a thirty-second live segment will vary. The traditional solution — a broadcast limiter on the studio chain — manages peaks but does not deliver consistent integrated loudness. This problem has no perfect solution; live audio is inherently variable. The question is how to handle the variability gracefully.

The content-type problem. An emergency announcement has different loudness requirements than a music host liner, which has different requirements than a news segment. Regulatory frameworks in some jurisdictions explicitly require emergency content to be audibly louder than surrounding programming — which creates a deliberate, sanctioned deviation from the normalized target. Advertising spots often have contractual or regulatory restrictions on their loudness level relative to programming. Music has its own normalization conventions that differ from spoken-word normalization because dynamic range in music content is a creative element, not a technical fault.

These three problems require different solutions, and an implementation that handles one well may handle another poorly.

How We Map the Standards to Our Production Environment

KAVANA's loudness processing lives in the MGR playout layer, applied per-segment before audio is committed to the playout queue, and at the transmitter output stage as a final safety net. The two-stage approach is intentional.

Pre-queue normalization handles the bulk of the loudness problem for pre-produced content. Every audio file that enters the playout queue goes through a loudness analysis pass that measures integrated LUFS, true peak, and the loudness range (LRA) using a BS.1770-4 compliant measurement implementation. Based on the measured values, the file is processed — gain adjustment, limiting, or transparent loudness normalization depending on the content type and the measured deviation from target.

For advertising spots, the target is -23 LUFS with a true peak cap of -1 dBTP, which matches the EBU R 128 specification. We process every spot on ingest, not at playout, so the processing overhead is not in the real-time path. A spot that measures at -17 LUFS integrated is gain-reduced by 6 dB before it enters the queue. The reduction is applied with a limiter pass to prevent true peak violations after the gain change — gain reduction alone does not guarantee that inter-sample peaks stay within the -1 dBTP limit.

For AI-synthesized content — news segments, weather, traffic, time calls — the normalization is applied in the synthesis pipeline rather than at ingest. We target -23 LUFS at the synthesis stage, which means the AI content arrives in the playout queue already normalized. The synthesis pipeline's loudness target is configurable per scene type, which allows us to set slightly different targets for different content categories while keeping everything within the BS.1770-4 tolerance window.

For music content, we apply a more conservative normalization: -16 LUFS with a true peak cap of -1 dBTP. This preserves more of the dynamic range that music production relies on and avoids the audible compression artifacts that full normalization to -23 LUFS can produce on music with wide dynamic range. The -16 LUFS target for music is common practice in streaming platforms; it is not the EBU R 128 target, but it is appropriate for the listening context.

The Transmitter Output Stage: Why a Safety Net Is Not Optional

Pre-queue normalization handles the predictable content categories. The transmitter output stage is there because some content is not predictable.

Live presenter audio from the studio arrives at the transmitter output stage without pre-processing. The AGC on the studio chain manages peaks but not integrated loudness. A presenter who reads into the microphone with significantly different energy levels in the first hour of the morning show and the third hour will produce audio that varies in integrated loudness by 2 to 4 LUFS, even with competent hardware processing on the studio chain.

The output stage limiter is not a loudness normalizer — it does not bring live audio to -23 LUFS integrated, because doing that for live audio requires a lookahead buffer that introduces processing latency, and latency in the live chain is audible and disruptive. What the output stage limiter does is enforce the true peak limit (-1 dBTP) and apply transparent gain reduction when the short-term loudness (measured over a 3-second window rather than the full integrated window) exceeds a configurable threshold. This prevents egregious loudness variations without producing the pumping and breathing artifacts of aggressive compression.

Networked content from provincial feeds presents a different problem: it arrives with processing already applied by the upstream facility, and applying additional normalization downstream can interact with the upstream processing in ways that produce artifacts — particularly with audio that has already been limited aggressively. For networked content, we typically measure on ingest and flag segments that fall outside the tolerance window for manual review, rather than applying automatic processing. Automatic processing on pre-processed material is often worse than leaving the material alone.

The Advertising Loudness Conversation Nobody Wants to Have

The loudness standards exist partly because of a documented pattern in broadcast advertising: spots produced as loud as possible to stand out, which triggered an arms race where every advertiser made their spots louder, which produced an experience where everything was uniformly loud and therefore equally fatiguing to listen to. The standards were intended to end the arms race by defining a maximum and normalizing everything to it.

In practice, the conversation at the station level is awkward. An advertiser who has paid a production house to make their spot sound big and present will not necessarily be happy to learn that the station is going to turn it down 6 dB before airing it. The processing is technically correct and regulatory-compliant, but the client may experience it as making their spot sound smaller.

We have encountered this at multiple stations. Our approach — which we offer to stations as an option, not as a requirement — is to provide advertisers with loudness analysis reports on their submitted spots as part of the spot approval workflow. An advertiser who receives a report showing that their spot measured at -17 LUFS and that the station will normalize it to -23 LUFS before airing can choose to have their production house deliver a version that was produced to target. This approach gives the advertiser control over the production quality of the normalized output rather than having normalization applied to a spot that was not designed for it.

The secondary benefit of this workflow is that it creates documented evidence of normalization application, which is relevant in regulatory contexts. If a regulatory body reviews the station's loudness compliance, having a log of every spot's pre-normalization LUFS measurement and the normalization applied is considerably more useful than having no records.

Loudness Across a Multi-Station Deployment: the Coordination Problem

Managing loudness at a single station is an engineering problem. Managing it consistently across a network of stations is an organizational and operational problem that happens to have engineering components.

For broadcasting groups that operate multiple stations — which describes most of our larger customers — the loudness question becomes: should normalization targets be consistent across all stations, or should individual stations be configured independently? There is a case for both.

Consistent targets across the network simplify content sharing. If a provincial broadcasting group produces a news package at headquarters and distributes it to fifteen county-level affiliate stations, having a single loudness target means the package does not need re-processing at each affiliate. Content produced to target at headquarters plays at target everywhere in the network.

Individual station configuration makes sense where stations have different transmission formats or different regulatory environments. An FM station in a city with specific local loudness requirements may need a different target than a medium-wave station in a rural county. A station that operates in a market with particularly aggressive advertising loudness norms — where clients are accustomed to louder spots — may need a different normalization threshold for advertising than a station in a market with more conservative norms.

The way we handle this in KAVANA is through a loudness configuration template that can be set at the network level and overridden at the station level. Content-type targets (programming, advertising, AI-synthesized, emergency) are configured separately, with the network-level template providing defaults and individual stations overriding only the parameters that differ for their specific situation. This gives groups the benefits of consistent defaults without forcing identical configuration on stations with genuinely different requirements.

The clock templates documentation at our GitHub organization includes loudness configuration examples for several common multi-station deployment patterns. These are starting points, not prescriptions — the right configuration depends on the specific content mix and transmission chain at each station.

The BS.1770-4 vs. R 128 vs. A/85 Difference: When It Actually Matters

For completeness, it is worth being explicit about where the standards differ and when the differences matter for our customer base.

BS.1770-4 defines the measurement algorithm. EBU R 128 and ATSC A/85 both use BS.1770-4 measurements but specify different target levels and different compliance windows. EBU R 128 targets -23 LUFS; ATSC A/85 targets -24 LUFS. The 1 LUFS difference is small enough to be inaudible but significant enough to be non-compliant with one standard if you are set up for the other.

For our Chinese customer base, neither EBU R 128 nor ATSC A/85 is formally mandated. The relevant Chinese broadcast standard for audio loudness is GY/T 287-2014, which defines target integrated loudness at -24 LUFS and true peak maximum at -3 dBTP — closer to ATSC A/85 than to EBU R 128, and with a more conservative true peak limit. Our default configuration targets -23 LUFS with -1 dBTP (EBU R 128) because this is the more common global reference point and because being 1 LUFS lower than the GY/T 287 target is compliant under that standard. Stations that want to configure strictly to GY/T 287 can set the target to -24 LUFS.

The true peak limit difference — -1 dBTP in EBU R 128 versus -3 dBTP in GY/T 287 — is more significant in practice. The -3 dBTP limit provides more headroom for inter-sample peak reconstruction and is a safer limit for content that will be encoded for digital transmission, where the encoding process can introduce additional peak levels. For stations feeding a digital transmission chain (DAB, digital cable, or IP streaming), the -3 dBTP limit is worth considering even if the station's primary reference is EBU R 128.

We default to -1 dBTP because most of our stations are FM or AM analog transmission, where the -1 dBTP limit is appropriate and provides slightly more loudness headroom. For stations with digital output, we recommend reconfiguring to -3 dBTP and have documentation in the MGR configuration guide that covers this adjustment.

What Good Loudness Practice Actually Sounds Like

The goal of loudness normalization is not to make everything sound the same. Music should sound like music, which means it has dynamic range and texture that news content does not have. An emergency announcement should sound different from a sponsor mention. The goal is to make every element of the programming sound as good as it can within a consistent level framework, so that listeners can adjust their volume once and trust that it will not need to be adjusted again.

When loudness normalization is implemented well, listeners do not notice it. They notice when it is implemented poorly — when a commercial break sounds dramatically louder than the programming, when music sounds compressed and lifeless because normalization has been applied too aggressively, when a transition between content types produces an audible level jump.

We have done a reasonable amount of listening testing across our station deployments to evaluate how the normalization implementation sounds in practice, and we are reasonably confident that the approach described here produces good results for the content mix our customers work with. For stations with unusual content mixes — very high proportions of live content, unusually dynamic music format, special regulatory loudness requirements — the configuration may need to be adjusted.

If you are evaluating loudness normalization for a multi-station deployment and want to talk through the specific challenges of your configuration, reach us at international@kavanafm.com. Loudness is one of those topics where the details matter and the generic recommendations do not always fit the specific situation.

KAVANA is developed by Hunan ShengGuang Technology Co., Ltd. (湖南声广科技有限公司), incorporated 2012, team active since 2005. We hold a broadcast production and distribution license (湘字第00565号) and operate under Chinese cybersecurity Level 3 certification. Technical documentation and open specifications: github.com/kavanafm.