11 Governing sound
The regulatory architecture for noise is not absent — it is elaborate. There are World Health Organization guideline values, a European directive with its mapping and action-planning machinery, and national and municipal instruments with limits and hours. And yet it fails the newest and fastest-growing sources, for a reason that has surfaced repeatedly since Chapter 2: the metric the whole edifice is built on cannot see them. A steady tonal hum from a data centre, or a heat pump on a party wall, can sit comfortably below an A-weighted, day-averaged limit while making a home unliveable. The failure looks exactly like compliance. And that failure is not merely a technical gap but something closer to a manufactured ignorance: noise kept, administratively, trivial. The map that was a way of knowing in Chapter 4 is here a way of governing, and of failing.
11.1 The architecture
Three layers sit, in principle, one above the other. At the top are the World Health Organization’s 2018 guideline values, the health-based floor: average levels recommended to stay below 53 decibels for road traffic, 54 for rail, 45 for aircraft and for wind turbines, with lower values again at night (World Health Organization Regional Office for Europe, 2018). They are the most evidence-driven numbers in the field, and they are recommendations. They bind no one. The recommendations are graded, strong for the transport sources and conditional for wind turbines, where the evidence is thinner. Graded or not, none is a limit. And each is expressed in the same averaged, A-weighted terms as everything beneath it. Even the health-based floor is laid on the very metric whose blind spots are at issue. They also carry a quieter limitation, since their categories are the familiar transport sources plus wind and leisure, and they contain no value at all for a data centre, a server hall or a domestic heat pump — the sources whose growth was traced in Chapters 6 and 7. What has no category cannot be exceeded.
Beneath them is the binding instrument of the European Union, the Environmental Noise Directive. It is genuinely binding, but on procedure rather than outcome. It requires member states to draw strategic noise maps and adopt action plans every five years, and it defines the indicators for doing so. But it sets no limit values of its own, prescribes no measures, and leaves both to national authorities, so that where limits exist and are breached, the directive supplies no penalty (European Environment Agency, 2025; European Parliament and Council of the European Union, 2002). And its reach begins higher than the harm does. The directive requires mapping only at and above 55 decibels for the day-evening-night average and 50 at night — thresholds that sit above the World Health Organization’s health-based values of 53 and 45. The band in which harm demonstrably begins is, for the purposes of the map that governs, below the horizon.1
The directive is not without ambition of another kind. Alongside the mapping, it asks member states to identify and preserve ‘quiet areas’ of good acoustic quality — a rare instance, in the whole apparatus, of a rule concerned with sound worth keeping rather than sound to be capped. But that clause is as unenforced as the rest, and the directive’s own review has conceded that national limits, where they exist, are often exceeded without consequence (European Environment Agency, 2025). A framework that maps diligently and compels nothing is what remains.
The layer with teeth is the national one. Spain’s noise decree sets binding quality objectives, among them a night-time limit of 55 decibels for a residential area. Unlike the layers above it, it does not evaluate a source by its plain A-weighted level alone. It adds correction terms: a penalty of up to six decibels where the sound carries an audible tone, and a further penalty for a marked low-frequency content (BOE, 2007). That detail will matter more than any other in what follows, because it is the one place in the whole architecture where the character of a sound, and not merely its averaged level, is written into the law. Municipal ordinances add the dimension of time to that of level — stricter limits by night, quiet hours, curfews on particular activities. It is at this local level, through the Granada ordinance and its kind, that the national decree is actually applied. But the local instruments inherit the national metric wholesale; they refine when and how loudly, never what is measured. So the architecture, read from top to bottom, is a recommendation without force, a duty to map without a duty to fix, and a binding limit that sees character only through a narrow and capped correction. Each layer is real; none, on its own, closes the gap.
11.2 The mismatch as a governance failure
Every layer of that architecture rests on the same foundation: the A-weighted, time-averaged decibel introduced in Chapter 2 (International Electrotechnical Commission, 2013) — the metric that, by construction, discounts the low frequencies the ear judges quiet and averages a steady tone into the background it hides in. The weighting was built to imitate the ear’s reduced sensitivity to low pitches at ordinary levels, and for a lorry passing in the street it serves well enough. For a sound that is all low pitch and never passes, it behaves almost as a filter designed to remove precisely the thing complained of (Leventhall, 2004). The consequence for governance is not incidental; it is structural. A source whose harm lives in the low, tonal, unremitting register can sit below every A-weighted limit while rendering a bedroom unusable — the data-centre drone, the compressor on the party wall. It is not that the rules are enforced too laxly. It is that the rules are met. The home is unliveable and the source is compliant, and both statements are true at once because the instrument that defines compliance cannot register what is making the home unliveable. These are not marginal cases. The sources that slip the frame most completely are the ones multiplying fastest — the data centres following the demand for computation, the heat pumps arriving with the energy transition, each of them steady, tonal, low and lawful (Chartered Institute of Environmental Health, 2026; Joint Legislative Audit and Review Commission, 2024). An architecture built for a twentieth-century soundscape of traffic and industry that rises and falls is being asked to govern a twenty-first-century one that hums without pause. It is the constancy as much as the pitch that the averaged number was never made to catch.
This is the 58-hertz tone of Chapter 7 returned in a new guise. There it disappeared from a measurement; here the same disappearance has legal force, because a harm the meter discounts is a harm the law cannot reach. The failure is built into the definition of the quantity being governed, which is why tightening enforcement, or lowering a limit, does nothing to fix it: an inspector can enforce an A-weighted limit to the decibel and still never see the hum. Several jurisdictions have noticed the gap and bolted separate low-frequency criteria onto their noise law (Moorhouse et al., 2005), and Spain’s correction terms are that same instinct written into the assessment itself. But almost everywhere these devices sit beside the binding number rather than inside it, as advisory guidance where the A-weighted figure is law, so that the recognition that something is missing has not yet displaced the quantity that misses it.
The picture is not uniformly bleak, and the exception is instructive. Spain’s correction terms are an attempt to make the instrument see character, and they can succeed. A data-centre hum sitting a few decibels below the plain limit is pushed above it once the tonal and low-frequency penalties are added, and becomes, at last, a breach that can be acted upon. But the correction is a patch on a frame it does not replace. The tonal penalty is capped (BOE, 2007); it is bolted onto a level still measured in A-weighted decibels; and a hum that lands just under the corrected limit is compliant by a single decibel, as untouchable as if the correction did not exist — the heat pump at fifty-four where the limit is fifty-five. The frame decides what the corrections can reach, and the frame is A-weighted. To see the harm reliably would take not a better patch but a different quantity. It is tempting to file all this under technical detail, a matter for acousticians to settle among themselves. But the choice of quantity is the policy. A limit is only ever a limit on something, and the something (here, the A-weighted, time-averaged level) decides in advance which harms can count as violations and which cannot. To choose the metric is to choose, before any limit is set or any inspector sent, the map of what will be governable at all.
11.3 Manufactured as trivial
There is a name in the history of science for ignorance of this kind. Agnotology — the study of how not-knowing is made — begins from the observation that ignorance is not always a simple void awaiting knowledge, but can be actively and structurally produced by how inquiry and its instruments are arranged (Proctor & Schiebinger, 2008, 2025). The concept is usually applied to the deliberate manufacture of doubt, and it must be handled here with more care than that, because the ignorance at issue is largely nobody’s plot. What does the producing, in the ordinary case, is not the suppression of a fact but the drawing of a category. A classification with no box for a kind of harm does not deny that harm; it has nowhere to put it, and a harm with nowhere to go is, in every register that counts administratively, absent.
It is manufactured all the same — not by conspiracy but by construction. The same pattern has been documented in climate governance, where the dominant instruments accept the existence of the problem while systematically placing its costliest implications outside the categories on which action depends (Lamb et al., 2020). The A-weighting curve, the averaging window, the fifty-five-decibel reporting floor, the list of sources that includes aircraft and omits server halls: each is a decision about what to render visible, and their sum is a decision, taken by no one in particular, about what to leave in the dark. A harm the official instrument cannot register is, administratively, not a harm. A source below the reporting threshold is, officially, not a problem. What is averaged away is not urgent; what has no category is not counted. The ignorance is real, it has effects, and it follows from the shape of the tools. The heat-pump complaint is the small, domestic form of it. A household kept awake by a tonal whine reports it; an officer measures an A-weighted level within the limit; the file is closed as compliant; and the record now shows not a harm endured but a complaint unfounded. The instrument has not merely failed to register the problem — it has produced a positive document attesting that there was none.
To notice this is not to allege bad faith. The regulators applying an A-weighted limit are doing their work correctly; the harm is not imaginary and the compliance is not fraudulent. The point is narrower and harder to dislodge: the shape of the instrument determines the shape of the ignorance, and a measurement gap treated as a neutral technicality quietly does a great deal of work. Whom that work serves is a fair question, provided it is not answered conspiratorially — the operator relieved of a nuisance the meter cannot find, the authority spared a problem it has no number for, the wider account of a digital and energy transition that would rather not carry this particular cost. None of them manufactured the blind spot. All of them, for now, inhabit it comfortably. To name the blind spot at all is faintly awkward, since it implicates instruments everyone relies on and a transition almost everyone endorses. It is easier, administratively and politically, to keep treating the hum as a private grievance than to concede that the public measure cannot hear it. That ease is itself part of how the not-knowing is kept.2
11.4 Comparing the regimes
The same disturbance can earn four different verdicts, depending only on which instrument is asked. Below, a single source at a bedroom window is put to all three regimes at once, alongside a fourth column for a metric attuned to how the sound is heard rather than to its averaged level. Switching between the road, the night flight, the nightlife terrace, the heat pump, the data-centre hum and the wind turbine changes which verdict falls where: a real disturbance rated compliant, or falling below the reporting floor, or slipping through for want of a category. One instrument, and one device within it, catches what the others miss.
Source · guideline values from World Health Organization Regional Office for Europe (2018); mapping and non-binding character from European Parliament and Council of the European Union (2002) and European Environment Agency (2025); binding limit and the tonal and low-frequency correction terms from BOE (2007) (applied locally through the Granada noise ordinance); A-weighting after International Electrotechnical Commission (2013). Source levels are illustrative, at the nearest façade at night. Accessed 1 August 2026.
11.5 What a metric that could see would change
The purpose of laying the failure out this way is not to arrive at a regulation to be passed, but to make plain what would come into view if the measurement changed. Diagnosis, not prescription, has been the discipline throughout. The changes are modest to state and consequential to make. Were the reporting floor lowered from fifty-five decibels to the health-based value beneath it, the band in which harm begins would appear on the map instead of below it. Were tonal and low-frequency character treated as a first-class quantity rather than a capped correction bolted to an A-weighted level, the data-centre hum and the heat pump would cease to be invisible, and would stop depending on a single national decree for their only chance of being seen. And were the governing quantity a measure of soundscape quality rather than of sound level alone, the instrument could begin to protect a quiet worth having, and not merely to set a ceiling on a loudness it can detect. The directive’s neglected clause on quiet areas is that idea already present, half-formed, in the law: it presumes that some acoustic environments are worth preserving, which is a claim about quality rather than level. What is missing is the metric to make the claim operational — a way of registering not how many decibels a place emits but what it is like to be within it.
That last shift is the largest, and it points beyond governance altogether. A regime built to cap a number can only ever say how much noise is too much; it cannot say what a place ought to sound like. The regulation at issue is not, in the end, absent or lax. It is precise about the wrong thing — exact to the decibel about a quantity that leaves out much of what does the harm. The distributional argument set out earlier compounds the point, for the harm the metric cannot see falls, like the rest, unevenly: the household beside the server hall or beneath the flight path is rarely the one with the means to litigate a hum or move away from it. What it lacks is not force but sight, and restoring that sight is less a matter of governing sound than of designing it.
Threshold logic is, by construction, a logic of population averages, and the sensitive tail is what such averaging omits. Migraine is the clearest case. Hypersensitivity to sound is not incidental to the disorder but part of its clinical definition: phonophobia is one of the criteria that separate a migraine attack from other headaches (Headache Classification Committee of the International Headache Society (IHS), 2018). The condition affects a large share of the adult population. Yet the case cuts two ways. The main population-based study to examine the question directly found no association between residential traffic-noise exposure and headache prevalence, only an indirect path running through noise-annoyance and disturbed sleep (Alkhalawi et al., 2021). A threshold set for the average is thus genuinely indifferent to a subgroup for whom ordinary sound can be aversive; but that indifference cannot itself be read as harm, because the chronic-exposure signal that would license the reading is, on present evidence, absent. What the metric renders invisible here is a distribution rather than a spectrum — the same selective blindness, displaced from the frequency axis onto the population.↩︎
The structural production of ignorance documented here should be distinguished from the active disinformation that accompanies extreme weather events, floods and heat waves, where misinformation attributing causation to non-climatic actors closes the windows in which attribution evidence is clearest (Daume, 2024). Both mechanisms keep evidence out of circulation, but through different means; conflating them would obscure the kind of reform each would require.↩︎