The Real Noise — what sound does to a body, and how to live quieter

Before we begin: how this study was made, and what it cannot tell you

This study was assembled from nine research fronts, each written independently and then handed to an adversarial reviewer whose only job was to break it. The reviewers broke a great deal. Four fronts had load-bearing errors — a WHO guideline value stated upside down, a blood-pressure figure overstated fourfold, a physics formula whose discrepancy pointed the wrong way, a meta-analysis credited with a finding it does not contain. Those corrections are logged at the end of this document, and they are not decoration. A study that asks other people to be careful has to show its own corrections first.

This document was later extended by five further fronts — on the standards history of A = 440, on what orchestras and traditions actually tune to, on which devices carry a tuning reference at all, on what is continuously emitted at global scale, and on what the Earth itself emits. Those fronts went through the same adversarial review, and it went badly for them too. One front's lead claim was found overstated on four separate counts and had to be rewritten from its first sentence; another had a documented finding reversed by its own cited source. Those corrections are in the log with the rest, and the new sections are 12 to 16.

It was then checked a third time, by a cold verification pass that was given no conclusions to defend and was not told what we had decided. That pass found three more things in this article, all recorded in a second corrections table at the end. Two of the three are omissions — evidence that would have made our own case stronger, left out. That is worth saying out loud, because it is the failure a careful study is most likely to have and least likely to notice: nothing on the page looked wrong.

Two limitations shaped everything that follows, and both are stated once, here, plainly.

The search budgets ran out. Several fronts exhausted their web-search allowance before every claim could be traced to a primary document. Where that happened, the claim is tagged unverifiable rather than quietly dropped or quietly kept. Unverifiable means we could not confirm it in the sources we reached. It does not mean it is false. We have worked hard not to let those two slide together, because letting them slide is the single thing this house does not forgive.

Two of our own fronts disagree with each other, and we have not resolved it. The European Environment Agency published a major report in June 2025. One front found evidence of a corrigendum incorporated in March 2026 that changed several health estimates; another front, reading a different set of EEA pages, found the original figures reproduced across four independent renderings and could not determine which version is authoritative. Both fronts are reporting what they saw. We show both sets of numbers below and tell you where the disagreement sits. We do not pretend to have settled it.

Claims are tagged as we go. [documented] means a source was actually read and says what we say it says. [contested] means there is a real scientific dispute, or the evidence is thin enough that a competent skeptic would win the argument. [myth] means popular and unsupported. [calculated] means arithmetic anyone can reproduce. unverifiable means we looked and could not confirm, and we are telling you so instead of filling the gap with confident prose.

One further thing, and it matters more than the tags. Nothing in this study is medical advice. Nothing here is a diagnosis, a treatment, or a recommendation to any particular person about their own body. What follows is a description of what has been measured across populations. Populations are not people. A finding that holds across two hundred thousand people says almost nothing about you specifically, and anyone who tells you otherwise is selling something.

1. The feeling, and why it is not imaginary

Start with the thing that brought you here.

You have probably noticed that the world is louder than it used to be, or at least louder than it should be. Not dramatically — not sirens and explosions — but continuously. A road that never fully quietens. A ventilation unit two buildings over that you only hear at 3 a.m. and then cannot stop hearing. An office where you can make out every third word of a conversation you are not part of, which is exactly the worst amount. Delivery vehicles reversing. The particular hum of a refrigerator in a small flat. And under all of it, a sense that you are being asked to concentrate, or rest, or recover, inside a room that will not let you.

Many people who feel this go looking for an explanation, and the internet offers them a very good story. The story says that the world's music was retuned in the twentieth century, from a warm and natural A=432 hertz to a harsh and imposed A=440, and that this is why everything feels wrong. It is a story with a villain, a date, and a fix. It travels beautifully.

The feeling is correct. The story is not the reason.

This is the whole argument of this study, and it is worth being precise about what is being said, because two very different things are often confused. We are not saying "you are imagining it, calm down." We are saying the opposite: the harm you are sensing is real, it is measured, it is large, and there is a serious scientific literature about it that almost nobody reads — and it has nothing to do with concert pitch. A companion study, slug 432-hz, takes the tuning narrative apart claim by claim and finds that two of its six component claims survive and four do not. That work is not repeated here. This study takes the instinct underneath it and puts it where the evidence actually is.

Here is the relocation in one sentence. The documented variables are loudness, timing — especially night — and chronicity. Not pitch. Not tuning reference. Not frequency in the sense the wellness internet uses the word. Every threshold that any health authority anywhere has issued about environmental sound is expressed as a sound pressure level in decibels [documented]. Not one is expressed in hertz. That is not an argument from silence; it is a statement you can verify by opening the recommendation tables of the World Health Organization's 2018 guidelines and looking at the units.

And the scale of what is documented is not small. The WHO's regional office for Europe, working with the European Commission's Joint Research Centre, published an estimate in 2011 that at least one million healthy life-years are lost every year to traffic-related noise in the western part of the European Region [documented]. That figure has problems, which we will examine honestly and at length, because it is quoted constantly and understood rarely. But even after every honest deduction, what remains is one of the largest environmental health burdens in the developed world, and it is treated as a nuisance rather than an exposure.

A person who arrives believing the 432 story and leaves knowing about night-time sound pressure levels has not been corrected. They have been armed.

2. What sound does to a body: the pathway

To understand why noise is a health exposure rather than an annoyance, you have to let go of one intuition: that the ear is the organ at risk.

The ear is at risk, at high levels, and we will come to that. But the ear is not where most of the documented damage sits. Occupational limits — the ones designed to stop factory workers going deaf — sit around 85 A-weighted decibels averaged over an eight-hour shift, with the permitted time halving for every 3 dB increase [documented — NIOSH recommended exposure limit]. The United States' National Institute on Deafness and Other Communication Disorders states that sounds at or below 70 dBA, even after long exposure, are unlikely to cause hearing loss, and that long or repeated exposure at or above 85 dBA can cause it [documented, quoting NIDCD directly].

Now compare: the WHO's recommended limit for road traffic noise is 53 dB Lden, and for night-time road noise 45 dB Lnight [documented]. That is thirty to forty decibels below the hearing-protection threshold. On a logarithmic scale, that is not a small margin — it is a different world of sound entirely.

Why would a guideline sit forty decibels below the level that damages the ear?

Because the ear is not the endpoint. The endpoint is the stress axis.

The proposed pathway, described as a model rather than a proof

The mechanism most cited in the cardiovascular literature is set out by Münzel, Kuntic, Daiber and Sørensen as a "noise reaction model" [documented as their model]. Its logic runs like this. Sound is processed continuously, including during sleep — the auditory system does not switch off. Sound that the brain categorises as unwanted or threatening activates limbic structures. That activation drives the hypothalamic-pituitary-adrenal axis and the sympathetic nervous system. Cortisol rises; catecholamines rise; angiotensin II, endothelin-1 and aldosterone rise. Downstream, NADPH oxidase (NOX2) is upregulated, generating reactive oxygen species which uncouple endothelial nitric oxide synthase. Nitric oxide availability collapses. The endothelium — the single-cell lining of every blood vessel you have — stops behaving properly. Blood pressure rises. Inflammation rises.

That is a specific, molecularly detailed chain, and it is one of the reasons this field is taken seriously by cardiologists. It is also, in its full form, a model [contested as a complete human pathway]. Pieces of it are well demonstrated. NOX2-deficient mice are almost completely protected from noise-induced vascular injury [documented in model systems]. Claims that circulate about circadian gene dysregulation (Bmal1, Cry1, Per2) and about the severity of injury in eNOS-knockout animals rest on review articles rather than on primary papers we were able to open, and are tagged unverifiable in this pass.

In humans, the demonstration is smaller and more careful. Schmidt and colleagues, publishing in the European Heart Journal in 2013, ran a randomised, blinded, crossover field study in 75 healthy adults in their own homes — a design detail that matters, because it is not a laboratory artefact [documented]. Participants were exposed to zero, thirty, or sixty simulated aircraft noise events at a mean maximum sound pressure level of 60 dB(A). Flow-mediated dilation — a standard measure of endothelial function — fell from 10.4% to 9.7% to 9.5% across the three conditions. The direct comparison reached P = 0.052, which is to say it did not reach conventional significance; the monotone dose-trend across all three conditions reached P = 0.020 [documented]. Morning adrenaline rose (P = 0.0099) [documented]. Sleep quality degraded, and pulse transit time shortened.

That study is frequently cited as proving the oxidative mechanism, because a sub-study found that two grams of oral vitamin C improved flow-mediated dilation. The vitamin C arm was five people [documented]. Five. A mechanistic conclusion cannot rest on five people, and this study will not let it. Tag: [contested]. The pathway is plausible and partially demonstrated; the specific oxidative-stress link in humans is suggestive and underpowered.

The part that is not in dispute

Strip out everything contested and something solid remains: the body responds to sound during sleep without the person waking, and the response is autonomic and endocrine rather than merely psychological [documented].

The clearest population evidence comes from HYENA, published by Haralabidis and colleagues in the European Heart Journal in 2008. One hundred and forty residents living near four European airports wore ambulatory blood-pressure cuffs measuring every fifteen minutes, with simultaneous noise measurement. An aircraft noise event was associated with a systolic blood pressure rise of 6.20 mmHg (95% CI 0.63 to 11.77) and a diastolic rise of 7.39 mmHg (3.09 to 11.69) [documented]. The effect was comparable across aircraft, road and indoor sources — the per-5-dB LAmax diastolic coefficients were 0.64 for aircraft, 0.55 for road and 0.68 for indoor noise [documented]. Not identical; comparable.

One correction our own reviewer forced, and it is instructive. HYENA is routinely cited — including in an earlier draft of this study — as showing that blood pressure rose in sleepers who did not wake. HYENA cannot show that. It used blood-pressure cuffs and noise meters. It had no EEG and no polysomnography [documented]. The claim that arousal is not required comes from laboratory work that the HYENA authors cite, not from HYENA itself. The finding stands; the attribution was wrong; we are saying so.

So the honest statement of the pathway is this. Sound at levels far below anything that threatens hearing produces measurable cardiovascular and endocrine responses, particularly at night, particularly when it is intermittent, and particularly when the person has no control over it. Whether that chain, sustained over decades, produces the disease burden attributed to it is the question the next four sections are about — and the answer is genuinely more complicated than either side of the argument usually admits.

3. The measured burden — and why the numbers move

This section is where most popular coverage of noise goes wrong, in both directions. It either quotes a large number as though it were a body count, or it notices that the number moves and concludes the whole field is invented. Neither is right, and the truth is more interesting than both.

What the headline number actually is

The WHO Regional Office for Europe, with the Joint Research Centre, published Burden of Disease from Environmental Noise in 2011. Its headline: at least one million healthy life-years are lost every year from traffic-related noise in the western part of the WHO European Region [documented].

Four things about that sentence are load-bearing and are almost always dropped.

It says traffic-related noise, not all environmental noise. [documented] Neighbour noise, industrial noise and leisure noise are not in it.

It says the western part of the European Region, not Europe, not the world. [documented]

It was published in 2011, using exposure–response functions that the WHO's own 2018 guidelines subsequently superseded for current work [documented].

It is a modelled estimate, not a count. [documented] No individual death or illness anywhere has been identified as noise-caused. The method is: modelled population exposure, multiplied by a pooled exposure–response function, multiplied by background prevalence, multiplied by a disability weight. Four modelled inputs stacked on each other, each with its own uncertainty.

The arithmetic problem, stated rather than hidden

The 2011 report gives per-outcome estimates. The two largest are firmly confirmed across every source we reached: sleep disturbance, 903,000 disability-adjusted life-years; annoyance, 587,000 DALYs [documented].

Three further figures circulate widely: ischaemic heart disease 61,000, childhood cognitive impairment 45,000, tinnitus 22,000. Our adversarial pass on the burden front could not confirm any of those three against a source it could reach, and flags them as unverified pending someone opening the report PDF by hand [unverifiable]. Other fronts of this study repeated them as documented. We are showing you the disagreement rather than picking the version that reads better.

Now the arithmetic. Those five components sum to 1,618,000 [calculated]. The headline says "at least one million." A hostile reader will call that an arithmetic failure, and they will say it loudly.

The standard explanation is that these are separate endpoint estimates rather than addends — different aggregation bases, with overlap between sleep disturbance and annoyance in the same individuals — so the headline is a conservative floor rather than a total. That explanation is coherent. It is also not stated in WHO's own text as far as we could verify [unverifiable]. We are offering it as the most likely reading, not as WHO's stated reasoning. An earlier draft of this study asserted the reasoning as though WHO had given it. That was our error.

And one more honest note, which cuts against the size of the headline: roughly 92% of that 1.618 million total is annoyance plus sleep disturbance [calculated]. Both are self-reported states. Neither is a diagnosed disease. Both carry disability weights that sit outside the standard Global Burden of Disease weight set and are genuinely contested [contested]. If you removed annoyance from the calculation entirely — and there are serious people who argue it does not belong in a DALY — the total would collapse by more than a third.

This is not a reason to dismiss the figure. It is a reason to understand what kind of figure it is. It is a policy-grade estimate of population-level welfare loss, not a mortality count.

How much the disability weights matter: a real demonstration

People sometimes wave at "modelling uncertainty" without showing it. Here it is, shown. Germany's national noise burden study explicitly calls disability weights "a sensitive modeling parameter" and demonstrates that raising the ischaemic heart disease weight to 0.266 doubles the estimated burden [documented]. One parameter choice, one factor of two.

An earlier draft of this study claimed specific variance shares — that annoyance disability weight accounted for roughly 30% of total variance, road fraction 32%, sleep disability weight 14%. Our reviewer checked the two obvious candidate studies for such a decomposition and found that neither performs one [documented absence in those two studies]. Those percentages are unverifiable as written and have been removed. The German doubling result is what we can actually show.

Current exposure: how many people, at what level

The European Environment Agency's Environmental noise in Europe 2025 draws on the 2022 reporting round under the Environmental Noise Directive (2002/49/EC). Its exposure findings [documented]:

Even the exposure counts move. The EEA's own exposure indicator page, for the same 2022 data, gives approximately 109 million total and 89 million for road [contested — numeric]. That is a spread of about 3% across the agency's own publications. Small, but worth knowing before you quote a figure to the decimal place.

What is not in dispute is the direction of travel. The European Union's Zero Pollution target is a 30% reduction by 2030 in the number of people chronically disturbed by transport noise, against a 2017 baseline. Between 2017 and 2022 the decline was approximately 3% [documented]. The target will not be met.

The health estimates, and our unresolved disagreement

Here is the point at which two of our own fronts diverge, and we are going to lay it out rather than choose.

Set A, which appears in the EEA's June 2025 newsroom release, the EEA's thematic briefing 3.3, a commentary in the European Heart Journal, and an independent peer-reviewed review published in 2025 — four separate renderings [documented]: 66,000 premature deaths; 50,000 new cardiovascular cases; 22,000 new type 2 diabetes cases; approximately 1.3 million DALYs; around 17 million highly annoyed; 4.6 million severely sleep-disturbed; children — approximately 560,000 with reading-comprehension impairment, 63,000 with behavioural problems, 272,000 with overweight; cost at least EUR 95.6 billion, about 0.6% of GDP.

Set B, which two of our fronts found on the EEA's own publication landing page, one of them describing it as the product of a corrigendum incorporated 27 March 2026 stating that an error had been identified in several numerical health estimates [documented as read, contested as to status]: 73,000 premature deaths; approximately 49,000 cardiovascular cases; 23,000 diabetes cases; 1.5 million DALYs; at least EUR 100 billion; more than 500,000 children with reading difficulties, 115,000 behavioural, 223,000 overweight.

A third front, reading the same landing page, saw Set B but found no corrigendum notice and no scope explanation, and concluded only that both sets are the EEA's own and the scope difference is simply unstated [unverifiable].

We are not able to resolve this. What we can say with confidence:

Both sets are EEA figures. This is not a case of one being a misquotation. The difference is roughly 10% in mortality and near-zero in cardiovascular cases. Anyone citing either should check the live EEA publication page before publishing, because at least one front of this study found evidence that the figures were revised, and a great deal of downstream coverage — including the EEA's own press release — may still carry the earlier set. An earlier draft of this study made a confident guess that the difference was EU-27 versus EEA-38 scope. No source we reached supports that guess, and it has been withdrawn [unverifiable].

Why the numbers jumped — and why that is the honest story

In 2020, using 2017 data, the EEA reported approximately 12,000 premature deaths and 48,000 new cases of ischaemic heart disease annually, with 22 million chronically highly annoyed and 6.5 million with chronic high sleep disturbance [documented]. It also reported 12,000 schoolchildren with reading impairment — and that figure was for aircraft noise specifically, not all sources [documented; an earlier draft of this study gave it as 12,300 and as all-source, wrong twice].

So: 12,000 premature deaths in 2020, 66,000 or 73,000 in 2025. A five- to six-fold change in five years. And 12,000 children from aircraft noise, to over 500,000 children from all sources.

A reasonable person seeing those two pairs of numbers concludes one of two things: either Europe got catastrophically louder in five years, or somebody is making it up.

Neither. The jump reflects changed exposure–response functions, added health outcomes that were not previously modelled, expanded coverage, and — per the corrigendum our front found — at least one arithmetic correction [documented]. It is a methodology change, not a measured worsening. Exposure over the same period fell by about 3%.

And that is the most important thing this section has to say. The volatility is the story. Any single mortality number from this literature should be read as model output with a wide and largely unstated uncertainty band. That is not a scandal — it is what health impact assessment is. But it means a person who quotes "66,000 deaths a year" as though it were a coroner's tally is overstating, and a person who notices the number moved and concludes the field is fraudulent is making the opposite error, equally badly.

The two objections a skeptic will raise, and our answers

"WHO and EEA are not independent witnesses." Correct, and this is the strongest technical objection in the section. The EEA computes its burden figures by applying WHO's exposure–response functions to modelled European exposure [documented — EEA states this]. Listing both as corroborating sources overstates the evidence base. If the WHO function shrinks, the EEA burden shrinks with it, automatically. They are one estimate wearing two coats.

"Traffic noise and traffic air pollution are the same exposure." This is the standing objection to every noise-cardiovascular figure ever published, and it is not answered by any of the numbers above. Road noise co-occurs almost perfectly with nitrogen dioxide and particulate matter, and both co-occur with poorer neighbourhoods and lower socioeconomic position. Cohort studies adjust for these, and the adjustment is imperfect. Residual confounding remains the unresolved threat to the entire burden estimate [contested]. Any account of this field that does not name it is not being straight with you.

One more structural caution. The exposure maps themselves are modelled, not measured — computed under the Environmental Noise Directive with known and uneven national reporting. The European Heart Journal commentary states that mapping excludes smaller airports, secondary roads and local railways, meaning the exposure counts are an undercount [documented]. An earlier draft of this study claimed the 2022 dataset was roughly 84% complete. No source we reached states any completeness percentage. The 84% has been deleted [unverifiable].

4. Sleep: the largest lever

If you read only one section of this study, read this one. Sleep disturbance is the single largest component of the estimated burden — 903,000 of the WHO's DALYs, more than half the itemised total [documented] — and it is also the component where an individual has the most leverage, the cheapest interventions, and the clearest physical picture of what is happening.

The dose that disturbs

The WHO's Environmental Noise Guidelines for the European Region (2018) issue strong recommendations to reduce night-time noise below [documented]:

Two technical points that are constantly misread.

Lden is not an average in the ordinary sense. It is a 24-hour average with penalties: +5 dB applied to the evening period (19:00–23:00) and +10 dB applied to the night (23:00–07:00) [documented]. The metric itself encodes the finding that night-time sound costs more. That is not a modelling convenience; it is the empirical result written into the unit.

These are outdoor façade values, annual-average, measured at the building's exterior. They are not bedroom levels [documented]. A great deal of confusion follows from people comparing a 45 dB guideline to a sound meter reading taken indoors.

The earlier Night Noise Guidelines for Europe (2009) set 40 dB Lnight outside as the guideline value, with 55 dB as an interim target for countries taking a stepwise path [documented]. An earlier draft of this study had those two inverted — presenting 55 dB as the guideline and 40 dB as the target. That is a serious error and it is corrected here. The often-quoted "about 30 dB inside the bedroom" figure comes from WHO's Community Noise Guidelines of 1999, not from the 2009 night document [documented].

The wind turbine entry deserves a note. WHO issued a conditional recommendation of below 45 dB Lden and declined to issue any night value at all, because it judged the evidence quality too low [documented]. That is a health authority saying, in public, that it does not know. It is worth admiring rather than glossing over.

How the thresholds were derived — and one thing that surprised us

The 2018 guidelines rest on eight commissioned systematic reviews — seven on health outcomes plus one on intervention effectiveness — using GRADE methodology explicitly adapted, in WHO's own words, "to the observational studies, which are usually the only source of research evidence in this area" [documented]. That adaptation is a real limitation and WHO states it themselves.

Basner and McGuire's sleep review screened 74 studies, of which 30 supplied usable data for source-specific exposure–response curves for the proportion of people "highly sleep disturbed" [documented]. Smith, Cordoza and Basner updated it in Environmental Health Perspectives in 2022, adding eleven studies and 109,070 survey responses, pooling 36. Where the survey question named noise as the source of disturbance, the odds of high sleep disturbance per 10 dB of Lnight were [documented]:

Those are large effects. And here is the honest limit that must travel with them: certainty fell to low or very low where noise was not named in the question [documented]. When you ask someone "how much does traffic noise disturb your sleep," you have told them what to blame. That is attribution bias, it is the standard critique of this literature, and it is the reason the certainty ratings differ. The update also carried its own caution in the other direction: populations exposed to high aircraft noise may be at greater risk than previously determined [documented].

An earlier draft of this study also asserted that the aircraft threshold sits eight decibels below the road threshold "because people are more disturbed by aircraft at identical levels." The subtraction is right; the causal explanation is not verified, and worse, the draft presented the aircraft recommendation as uncontroversial when it has a published dispute — Gjestland challenged the basis for WHO's aircraft annoyance recommendation in IJERPH in 2018, and a published Comment replied [documented]. That was the most attackable line in the draft. It is corrected: the aircraft value is contested in print, and the reason for the gap is unverified.

What happens without waking

This is the part that makes noise a health exposure rather than a comfort question.

Awakening is not the injury. It is one visible symptom of the injury.

Physiological reactions — hormonal secretion, cortical arousals, body movement — have been reported at levels "probably around 33 dB," with awakenings from around 48 dB. That figure comes from Halperin's 2014 review, summarising others, and the source itself hedges with "probably" [contested — no single universal threshold is established, and the 33 dB floor outside a sleep laboratory is unverifiable].

The HYENA blood-pressure findings described in section 2 belong here too: sound events raising systolic pressure by around 6 mmHg and diastolic by around 7 mmHg, in people asleep at home [documented], with the caveat already stated that HYENA had no EEG and cannot itself prove the sleeper stayed asleep.

Architecture: what is actually being taken

Intermittent night noise does something specific and measurable to the shape of sleep. It fragments it, suppresses N3 — slow-wave sleep, the deepest stage — and shifts that time into lighter N2 [documented].

The cleanest demonstration is recent. Basner and colleagues published a randomised crossover polysomnography trial in SLEEP in 2026 (49(5):zsag001). Twenty-five healthy adults, mean age 28.5, seven of them male, slept seven nights each in a laboratory. On exposure nights they received 93 recorded traffic events at maximum levels of 45 to 65 dBA, against a noise-free control night at an ambient LAeq of 23.7 dB. Results [documented]:

N3 fell by 23.4 minutes (p < .0001). N2 rose by 20.8 minutes (p = .0018).

Roughly twenty-three minutes of the deepest sleep, converted into lighter sleep, every exposed night. The participants were young, healthy, and in a laboratory; the sample was twenty-five; and no clinical outcome is attached to those 23 minutes [contested]. What it demonstrates is the mechanism at the level of sleep architecture, not the disease consequence.

The habituation trap

Almost everyone who lives on a busy road will tell you they got used to it. That report is sincere and it is also, in a specific and important way, unreliable.

Basner, Müller and Elmenhorst, in SLEEP in 2011 (34(1):11–23), studied 72 subjects and found that cortical arousals and sleep continuity showed habituation across nights, while cardiac arousals did not [documented]. The conscious, reportable part of the response adapts. The cardiovascular part keeps responding.

An earlier draft cited this as a 2008 paper and described the finding as habituation "within the night." Both wrong; corrected. The design detail sometimes attached to it — four nights weekly across three weeks at 45–77 dBA — is unverifiable; the abstract we reached describes eleven consecutive nights.

The consequence for the whole field is large: self-report is a poor instrument for this exposure. A survey asking "does noise bother you?" systematically under-detects, because the population most exposed is the population that has stopped noticing consciously. That cuts against the noise literature's own annoyance-based DALY estimates in one direction, and for the seriousness of the exposure in another. Both are true at once.

What helps, and where the evidence thins fast

Earplugs — the best-evidenced personal measure, with a ceiling. In the same 2026 SLEEP trial, earplugs mitigated nearly all the environmental-noise effect, recovering 16.9 of the 23.4 lost N3 minutes — about 72% [documented]. And they began failing at 65 dBA [documented]. That is a genuinely useful, honest result: effective across the ordinary range, and not a solution to a genuinely loud environment.

Whether earplugs help in intensive care, and whether they affect delirium, is unverified here. An earlier draft asserted "no consistent delirium benefit." We could not confirm that, literature exists pointing the other way, and the claim has been withdrawn rather than kept.

White and pink noise masking — weaker than its popularity suggests, and pointing partly the wrong way. This deserves care, because it is the single most common thing people do about noise, and the evidence is not what they think.

Riedy and colleagues, in Sleep Medicine Reviews in 2021, reviewed 38 articles and rated the evidence that continuous broadband noise improves sleep as very low quality by GRADE, too heterogeneous to pool, and found continuous noise both to improve and to disrupt sleep depending on the study [documented]. An earlier draft of this study said continuous noise "tended to shorten sleep onset and reduce fragmentation." That imports a direction the review does not support, and it has been corrected.

Ding and colleagues, in Sleep Medicine in 2025, pooled 12 randomised trials with 1,301 participants and found improved Pittsburgh Sleep Quality Index scores in adults (mean difference −3.70) and older adults (−2.71), and in infants and toddlers a 12-hour sleep-efficiency gain of 6.62% with no effect on 24-hour efficiency [documented]. Note carefully what those adult findings rest on: PSQI is self-report — the exact instrument the habituation research shows is unreliable for sound.

Against it, an objective measure: pink noise at 50 dBA cut REM sleep by 18.6 minutes versus quiet (p = .0003) [documented]. The authors explicitly cautioned against widespread indiscriminate broadband-noise use, citing REM sleep's role in memory, emotion regulation and neurodevelopment [documented].

And there is a documented harm signal at the loud end that parents in particular should know exists. Hugh, Wolter, Cushing and colleagues, in Pediatrics in 2014, measured fourteen infant sleep machines: three of fourteen exceeded 85 dBA at 30 cm at maximum volume, and all fourteen exceeded 50 dBA, the recommended nursery limit [documented]. An earlier draft of this study claimed a peer-reviewed measurement above 91 dB; we could not locate it, and it has been replaced with the figure that is verifiable.

So the honest position on masking: it is popular, its subjective evidence is very low quality and bidirectional, its one clean objective finding is a REM reduction, and at high volumes close to an infant it is itself an exposure. That is not a prohibition. It is a description of what the evidence supports, which is considerably less than the marketing.

Source and building control — quieter road surfaces, night flight restrictions, façade and glazing treatment, bedroom orientation away from the road — is what the entire WHO and EEA guidance architecture exists to drive [documented]. Individual masking is the weakest link in the chain, not the strongest. The specific figures often quoted for these measures — barriers giving 6–7 dB in the 500 Hz to 1 kHz band, a "quiet façade" threshold of 10 dB(A) or more, the sleep benefit of bedroom reorientation — are unverified in this pass and are named here as directions supported by guidance rather than as quantities we can vouch for.

Honest gaps in this section. Long-term randomised trials of any night-noise intervention against hard cardiovascular endpoints: none identified [unverified absence — we did not search to exhaustion, and an absence we did not exhaustively search is not a documented absence]. Whether nightly earplug use over years is net-beneficial: unverifiable. Nothing here is guidance for any individual's sleep.

5. The heart: the front where the evidence got weaker, not stronger

This is the section most likely to disappoint a reader who arrived wanting ammunition, and it is the section this study is proudest of.

Between 2018 and 2025, the estimated cardiovascular effect of road traffic noise shrank by roughly three-quarters. Not to zero. But the confident "8% more heart disease per 10 decibels" that underpins most public communication about noise is superseded, and honest advocacy has to say so.

What WHO used in 2018

The 2018 guidelines relied on van Kempen and colleagues' systematic review, which pooled seven longitudinal studies to a relative risk of 1.08 (95% CI 1.01–1.15) per 10 dB Lden for road traffic noise and incident ischaemic heart disease, graded high under GRADE [documented].

Two things about that are routinely dropped, and both matter.

In the same review, hypertension was graded very low and stroke was graded low. [documented] Ischaemic heart disease was the outlier in evidence quality, not the pattern. Citing "GRADE high" without saying that is cherry-picking, and an earlier draft of this study did exactly that.

The 53 dB Lden guideline value is not a cardiac threshold. [documented] It derives from the annoyance exposure–response curve, at the point where 10% of the population is highly annoyed. van Kempen states that cardiac risk "increases continuously for road traffic noise levels from about 50 dB Lden," and designated no threshold at all. An earlier draft of this study said risk "was assumed to rise linearly from 53 dB." That was a conflation of two different curves. Correcting it actually strengthens the argument rather than weakening it: it shows that WHO's most-quoted number is an annoyance threshold being widely read as a cardiac one.

What the newer meta-analyses found

Pershagen and colleagues, in Environmental Epidemiology in 2025, pooled 20 publications covering 28 study populations and more than 8.4 million people. Per 10 dB Lden of road traffic noise [documented]:

Read that table carefully, because it is routinely misread in both directions.

An earlier draft of this study reported the 1.017 figure as though it applied to "IHD, MI and stroke" collectively, and used it to describe the cardiac finding as null. That is a cherry-pick, and it is the error our reviewer hit hardest. The IHD estimate alone is null. Myocardial infarction and stroke are both significant, in larger pools, and the authors' own stated conclusion is that road traffic noise does increase the incidence of ischaemic heart disease, myocardial infarction and stroke [documented].

The correct statement is: the magnitude was revised down — from roughly 8% to roughly 2% per 10 dB — while the existence and direction were not overturned.

A second point of honesty. Restricting the analysis to studies with low exposure-assessment bias — the obvious repair, if measurement error is the explanation — raised the estimates but did not rescue IHD: IHD 1.026 (0.980–1.074), still crossing null and with a wider confidence interval; MI 1.040 (1.031–1.049); stroke 1.031 (1.014–1.048) [documented]. Splines showed risk rising with exposure for all three, with significant deviation from linearity for IHD and MI [documented].

The authors attribute the shrinkage largely to exposure-assessment bias: 13 of 22 studies (59%) were rated high risk in the paper's own table, while a comparative statement in the same paper puts 79% of the newer meta-analysed studies at high risk against 33% in the WHO-era set. Those two figures are inconsistent within the paper itself; both are reported here [contested].

The other current estimates

Three further values are in play, and giving only one of them would be dishonest [documented values, contested as a set]:

The honest framing is therefore a spread of roughly 1.7% to 8% excess risk per 10 dB — not a two-sided dispute between WHO and one dissenting paper, which is how an earlier draft of this study presented it. At the low end of that spread, the cardiovascular share of the total burden falls substantially.

Minkin and colleagues at the UK Health Security Agency published in Environment International in 2025, per 10 dB Lden [documented]: heart disease incidence, road 1.02 (1.01–1.04), all sources 1.03 (1.01–1.04); heart disease mortality, all sources 1.03 (1.02–1.05), aircraft 1.07 (1.01–1.14). Their exposure–response curves start between 32 and 45 dB Lden depending on source and outcome — well below WHO's anchor. Those are model-fit start points, not established no-effect levels, and aircraft estimates attenuated when small-area studies were excluded. An earlier draft of this study described the 1.03 figure as all-cause mortality. It is heart-disease mortality; the paper reports no all-cause figure. Corrected.

The outlier that should be named as an error

Tabaei and colleagues published an umbrella review of 20 meta-analyses in BMC Cardiovascular Disorders in 2025, reporting IHD 1.06 (1.03–1.09), coronary heart disease 1.08 (1.04–1.13), stroke 1.04 (1.00–1.08), atrial fibrillation 1.05 (1.02–1.09) — and hypertension 1.81 (1.51–2.18), stated verbatim as "per 10-decibel (dB) increase" [documented].

An 81% increase in hypertension per 10 decibels is irreconcilable with every cohort dataset in existence, which cluster around 1.02 and often cross null. An earlier draft of this study speculated that it was "almost certainly not a per-10-dB figure" — which was itself an unverifiable guess about someone else's paper. The honest handling: it is stated per 10 dB, it cannot be reconciled with the cohort evidence, and it should be treated as an error within the umbrella review. The mechanism of the error is unverifiable from the abstract. We are not going to invent one.

Hypertension: the weakest of the three

Design determines the answer, which is itself the finding [contested]. Cross-sectional studies pool to 1.05 (1.02–1.08) per 10 dB, with around 9% higher odds in some analyses. Cohort and case-control estimates are null — 1.018 (0.987–1.049) and 1.02 (0.98–1.05) [documented]. An earlier draft of this study gave "3–5% per 10 dB" as the hypertension figure. Too high, and it obscured the design dependence, which is the real story.

The human studies people cite, with what is wrong with each

Osborne et al., European Heart Journal, 2020. 498 adults who had undergone ¹⁸F-FDG PET/CT imaging; major adverse cardiovascular events hazard ratio 1.341 (1.147–1.567) per 5 dBA — note, per five decibels, not ten — with the effect mediated serially through amygdala metabolic activity and arterial inflammation, adjusted for air pollution and neighbourhood income [documented]. This is the most mechanistically elegant human study in the field, and it is also single-centre, retrospective, and 84.3% of the participants had a prior malignancy [documented]. That is not a general population. The authors state it cannot establish causality. An earlier draft presented it as clean. It is not clean; it is suggestive.

Saucy et al., European Heart Journal, 2021. A case-crossover study of 24,886 cardiovascular deaths near Zürich airport, 7,641 of them nocturnal, testing aircraft noise in the two hours before death. All cardiovascular deaths: OR 1.44 (1.03–2.04) above 50 dB Lmax, 1.33 (1.05–1.67) at 40–50 dB, trend p = 0.01 [documented]. Three cautions travel with it: the upper confidence bound nearly touches null; ischaemic heart disease specifically was not significant (trend p = 0.18), with the signal sitting in heart failure and arrhythmia; and the attributable fraction was around 3%. Zürich has a night curfew, so absolute exposures were low.

Schmidt et al., 2013 — described in section 2. Seventy-five people, field study, dose-trend p = 0.020, direct comparison p = 0.052, vitamin C arm of five.

Herzog et al., 2019, reporting similar findings with railway noise plus raised 3-nitrotyrosine and 8-isoprostane, is cited everywhere at n = 70. That sample size is unverified — it is sourced only to a review article which also misstates Schmidt's sample as 70 when it was 75 [documented discrepancy].

The causal verdict, stated honestly

Applying the Bradford Hill considerations, and refusing to flatter the field:

Met: biological plausibility, and unusually specifically — a named molecular pathway rather than a hand-wave. Temporality. Consistency across three independent kinds of evidence: epidemiology, human experiment, animal model.

Not met, and these are not small:

(a) No intervention trial has ever tested whether reducing noise reduces cardiovascular events. [documented] Not one. This is the decisive gap in the entire field, and it should stay prominent rather than buried in a limitations paragraph. Everything above is observational.

(b) The dose–response relationship is weaker than it was in 2018, not stronger. Fields that are converging on a true effect usually tighten. This one loosened.

(c) Independence from air pollution is not established. This is the correction that inverts an argument an earlier draft of this study made. That draft listed air-pollution independence as a Bradford Hill criterion met. Pershagen and colleagues state that adjustment for PM2.5 consistently reduced road traffic noise risk estimates, with only the stroke excess surviving, across roughly half the studies that permitted the analysis [documented]. The criterion is not met. It is contested. Our own key source said so and we had written the opposite.

The dominant residual threat runs the other way and partially rescues the field: exposure misclassification. Façade-level modelled Lden ignores window orientation, glazing, which room a person sleeps in, and whether they open the window. That kind of error biases estimates toward the null, and may explain part of the shrinkage [contested].

Münzel and colleagues themselves frame noise as a modifiable risk factor, not a proven cause [documented]. So does this study.

What this front establishes: the instinct that the modern acoustic environment injures the body is supported, at a smaller effect size than the headline literature implies, with the strongest evidence for myocardial infarction and stroke rather than ischaemic heart disease, and the weakest for hypertension. The operative variables are sound pressure level, timing, and chronicity. Nothing here is frequency-standard-dependent.

6. Low frequency and infrasound, fairly

This section is written for a specific person: someone who has complained to a council or a landlord about a hum, been told the measurement is compliant, and been left with the clear implication that they are imagining it.

That implication is often wrong, and the reason it is wrong is technical and demonstrable. It is also true that the popular explanation for such experiences — infrasound, secret frequencies, sub-audible weapons — is not supported. Both of those statements are in this section, and neither cancels the other.

Definitions, and why the boundary is a convention

Low-frequency noise is conventionally 20 to 200 Hz. Infrasound is below 20 Hz [documented]. The boundary is a convention, not a physical wall: audibility below 20 Hz depends on level [documented].

Infrasound is not inaudible — and that is the first thing that goes wrong in the popular story

Leventhall's review for the UK Department for Environment, Food and Rural Affairs (2003) states plainly that "frequencies down to a few hertz are audible at high enough levels" [documented]. Thresholds, from Watanabe and Møller (1990) as reported in that review:

At 20 Hz the ear is about 75 decibels less sensitive than at 1 kHz [documented]. An earlier draft of this study said "about 80 dB"; corrected to 75–76.

Two consequences follow, and they pull in opposite directions.

Against the popular story: environmental infrasound at residential levels is typically far below these thresholds. Something inaudible at 90 dB SPL is not sneaking up on anyone at 40.

For the complainant: the equal-loudness contours compress as frequency falls. An 80-decibel dynamic span at 1 kHz becomes a 40-decibel span at 20 Hz, and a doubling of perceived loudness takes about 5 dB rather than the familiar 10 [documented]. In plain terms: at low frequencies, a very small change in level produces a very large change in how loud something feels. An earlier draft of this study said "a few dB takes you from barely detectable to loud" and gave 10 dB for a loudness doubling. The corrected numbers are 40 dB of span and about 5 dB per doubling — and they make the point better.

The A-weighting problem, stated with its counter-example

Almost all noise regulation uses A-weighting, dB(A), which was designed to approximate human hearing sensitivity. It attenuates 70 dB at 10 Hz [documented]. A sound with substantial energy down there is, by construction, nearly invisible to a dB(A) meter.

The psychoacoustic consequences, all as reported in the DEFRA review [documented, small samples]:

At equal dB(A), noise dominated by 15–50 Hz energy was judged 4–7 dB louder and 5–8 dB more annoying (Kjellberg et al. 1984, n = 20). For broadband low-frequency noise, the annoyance underestimate was around 3 dB at 65 dB(Lin) and 6 dB at 70 dB(Lin) (Persson & Björkman 1988; Persson et al. 1990).

And the counter-example, from the same review, which an earlier draft of this study omitted: Landström et al. (1993) found dB(A) over-rated the effects of a 100 Hz noise band by 10 to 15.5 dB [documented].

So the correct claim is not "dB(A) systematically under-reports low-frequency noise." It is: dB(A) under-reports for tones and for broadband low-frequency-dominated noise, and over-reports in at least one documented band case. It is not uniformly conservative, and that caveat must travel with the claim.

What survives, and it is the operative point for the complainant: a compliant dB(A) reading can coexist with a real low-frequency noise problem [documented]. The officer with the meter is not lying, and neither is the resident.

The United Kingdom's response to exactly this is NANR45 (Moorhouse, Waddington and Adams, University of Salford, 2005; revision 1, December 2011): a third-octave criterion curve running from 92 dB at 10 Hz down to 34 dB at 160 Hz, with a 5 dB relaxation for daytime-only noise and a further 5 dB relaxation for steady, non-fluctuating sounds [documented]. An earlier draft of this study omitted the second relaxation and misdescribed the authorship as a three-author paper — it is a report for DEFRA by Dr Geoff Leventhall, assisted by Dr Peter Pelmear and Dr Stephen Benton, that being the DEFRA review; NANR45 is the separate Salford procedure. NANR45 is a complaint-assessment procedure, not a statutory limit [documented]. There is no UK exposure standard for low-frequency noise. ISO 7196 defines G-weighting for infrasound [documented].

Physiology and performance at low frequency

Persson Waye and colleagues, in Life Sciences in 2002 (n = 32), ran a two-hour task under ventilation noise at 40 dB(A) — quiet by any regulatory standard. Low-frequency-dominated noise produced longer response times and higher annoyance than a flat spectrum at the same level [documented]. On cortisol, the paper is careful and so are we: cortisol "was not significantly modulated by the noises or related to noise sensitivity alone," but the normal circadian decline "was significantly attenuated in subjects high-sensitive to noise in general" [documented]. That is a subgroup interaction in one laboratory with one biomarker and 32 people, never replicated at scale [contested].

Baliatsas and colleagues, in Science of the Total Environment in 2016, systematically reviewed seven observational studies and found associations with annoyance, sleep problems, concentration difficulty and headache [documented as association]. All seven are cross-sectional with self-reported exposure. The review is best read as a map of the literature's weakness rather than as evidence of harm [contested].

Wind turbines: the case study in doing this fairly

Health Canada studied 1,238 participants, with outdoor wind turbine noise reaching a maximum of 46 dB(A), and found "no apparent pattern or statistically significant relationship" between turbine noise and self-reported or actigraphy-measured sleep. Annoyance did rise with level (Michaud et al., Sleep, 2016) [documented]. Note the ceiling: 46 dB(A) is not a test of the loudest exposures.

Australia's NHMRC (2015) found no consistent evidence of adverse health effects from a small and poor-quality direct evidence base [documented — the often-quoted screening counts of "over 4,000 records, 13 of direct relevance" are unverified in this pass].

In the laboratory: turbine noise at 25 dB(A) produced no polysomnographic or diary sleep effects (Liebich et al., Sleep, 2022, n = 68), with the authors' explicit caveat that higher levels are untested [documented]. An earlier draft of this study described a Flinders study as showing infrasound "not audible awake, no cortical response." We could not locate that study. The reachable Flinders paper tested audible turbine noise at 25 dB(A) against a 19 dB(A) quiet night. It is not an infrasound null, and describing it as one was wrong.

And a genuinely surprising result, in a randomised trial at 30, 40 and 50 dBA (Lechat et al., Journal of Sleep Research, 2026, n = 62): wind farm noise onset was marginally less arousing than road traffic noise at 40 dBA and above — 2.1% versus 3.2% at 40 dBA, 5.0% versus 8.6% at 50 dBA [documented].

Against all that, the largest single study pointing the other way: Poulsen and colleagues, in Environmental Health Perspectives in 2019, a Danish register cohort of about 584,000 adults. The one statistically significant estimate: sleep medication redemption among people aged 65 and over exposed to night-time outdoor turbine noise of 42 dB or more, hazard ratio 1.68 (1.27–2.21) [documented]. Everything around it was null: antidepressants in that same group 1.23 (0.90–1.69); overall sleep medication 1.14 (0.98–1.33).

And the detail that matters most for this section: indoor low-frequency turbine noise showed no overall association [documented]. That result cuts against a low-frequency mechanism, and it comes from the study most often cited as evidence for turbine harm. An earlier draft of this study presented Poulsen as broadly supportive without any of these qualifications.

The measurement dispute, unresolved

Baumgart, Fritzsche and Marburg, in the Journal of Sound and Vibration in 2021, reanalysed a widely cited 2017 measurement of wind turbine low-frequency emission and reported levels "about 34 dB below the original work," with all measured levels at 200 metres "well below the hearing threshold" [documented]. Part of the discrepancy is a power-spectrum versus power-spectral-density normalisation issue and a power correction, not simply a mis-measurement.

An earlier draft of this study said a corrigendum was published in 2021, correcting the figure by 36 dB. No corrigendum exists. What exists is a Comment (Baumgart et al., 2021), an author Reply (Pilger & Ceranna, JSV, September 2022), and a further comment (Holzheu & Hundhausen, JSV, January 2024) [documented]. This is a live and unresolved dispute, not a settled correction, and the figure is 34 dB from the primary source, not 36. Claims resting on the original 2017 figure are correspondingly weakened; the correction does not settle the annoyance question either way.

Mechanism claims, sorted

Real observation: Salt and Hullar, in Hearing Research in 2010, showed that cochlear outer hair cells respond to infrasound at levels below inner-hair-cell detection [documented]. This is a genuine physiological finding and it is the strongest card in the infrasound hand.

Contested extension: that this outer-hair-cell response produces the symptom clusters attributed to it. No pathway has been demonstrated [contested].

The study everyone cites, in context: Weichenberger and colleagues, PLoS ONE 2017, n = 14, found altered resting-state brain connectivity from 12 Hz stimulation presented "near threshold" — which in that paper means 2 dB below each individual's threshold, where the mean threshold was 86.5 dB SPL [documented]. So roughly 84 dB. That is enormously above any residential environmental infrasound. The authors' suggested link to sleep disturbance, panic and depression is their speculation, not a result of the study [contested].

The countervailing trial, which an earlier draft omitted: Ascone and colleagues, Scientific Reports 2021, n = 38, exposed participants to 6 Hz at 80–90 dB across 28 nights and reported "null findings emerged for all behavioral variables" [documented].

[myth]: "vibroacoustic disease" caused by environmental infrasound. A bibliometric analysis found the claim propagated largely through author self-citation (Chapman & St George, Australian and New Zealand Journal of Public Health, 2013), with a rebuttal letter from the originating authors in 2014 [documented]. A frequently cited Health Protection Agency (2010) statement on the topic is unverified here — the report PDF returned 404.

Expectation, and the respect it is owed

In a double-blind trial with 72 participants and 23-minute exposures, baseline concern predicted symptoms while actual infrasound exposure did not (Tonin, Brett and Colagiuri, 2016), consistent with Crichton and colleagues' earlier work [documented].

Two things must be said about this, and the order matters.

First: nocebo symptoms are physically real. A headache produced by expectation is a headache. Nausea produced by expectation is nausea. Nothing about the word "nocebo" makes a person's suffering fictional or their honesty suspect. Rule 5 of this house applies with full force here: truth gently, and never mock someone for believing something that turned out not to hold.

Second: these are the weakest studies in the section, and we are going to say so ourselves rather than wait to be told. Twenty-three minutes of exposure in healthy volunteers does not model years of residential exposure. The nocebo literature is the strongest thing the skeptical side has and it is also the flimsiest thing on the table. It does not license telling a person who has lived with a hum for six years that they invented it.

The Hum

Documented cases where an identified source was found have generally turned out to be ordinary industrial machinery [documented, case by case]. The most-studied instance, the Windsor Hum, is instructive precisely because the two commissioned teams disagreed: the University of Windsor team concluded that blast-furnace operations on Zug Island were the "likely source" while conceding that "conclusive evidence of the source was not achieved"; the Western University team reported that the array bearing pointed "well to the South of Zug Island" and raised quarry activity or industrial ventilation as alternatives [documented]. An earlier draft of this study presented the two investigations as converging. They did not.

The frequently cited 1993 Taos Hum four-institution investigation is unverified in this pass. Deming's estimate that around 2% of the population perceives a Hum, and his radio-demodulation hypothesis (Journal of Scientific Exploration, 2004), are [contested] — prevalence unverified, mechanism unsupported.

The bottom line of this section

The documented harm mechanism at low frequency is audible low-frequency noise. Under-reported by dB(A) in the tonal and broadband-LF cases, disproportionately annoying because of loudness compression, and produced by entirely ordinary industrial and building plant — ventilation, compressors, pumps, transformers, extract fans. That is a real, actionable, unglamorous problem, and it is the one that most complainants actually have.

Whether sub-audible infrasound at real-world environmental levels causes any health outcome is unverifiable. The null is better supported. Few studies are powered to detect small effects. Whether Hum sufferers resolve after source mitigation: unverifiable. Whether the Danish over-65 signal is causal or residual confounding: open.

7. The cognitive cost: what noise takes from thinking

The strongest evidence in the entire noise field is not about hearts. It is about children reading.

Children

The RANCH study, published by Stansfeld and colleagues in The Lancet in 2005, remains the anchor. It studied 2,844 children aged 9 to 10 at 89 schools around Heathrow, Schiphol and Barajas airports. Chronic aircraft noise exposure showed linear exposure–effect associations with impaired reading comprehension (p = 0.0097) and impaired recognition memory (p = 0.0141), surviving adjustment for mother's education, socioeconomic status, longstanding illness, and classroom insulation [documented].

RANCH's own published conversion: an increase of 5 dB LAeq,16h in aircraft noise corresponds to approximately a 2-month delay in reading age in the United Kingdom and a 1-month delay in the Netherlands [documented].

An earlier draft of this study stated that a 20 dB school-to-school difference equals about two months by one conversion and about eight months by another. That is simply wrong. It garbled the published 5 dB figure and then presented the UK and Dutch results as two competing readings of one quantity when they are two different countries. Corrected above.

And note what the country difference itself tells you: the same decibel meant twice as much reading delay in the UK as in the Netherlands. Converting decibels into months of schooling is a modelling choice that does not travel cleanly across borders [contested].

An earlier draft also said RANCH found "no evident threshold below which exposure was safe." That is overreach. Linearity was observed within the exposure range studied. The absence of an observed threshold inside a range is not evidence that no threshold exists [contested].

The inconvenient result in the same paper, which any honest account must include: road traffic noise showed no equivalent reading effect, and was associated with increases in episodic memory — conceptual recall p = 0.0066, information recall p = 0.0489 [documented]. That is not a benefit of traffic noise. It is a marker of residual confounding in the study, and it costs nothing to disclose and everything to hide, because a hostile reader will find it first.

Clark and colleagues (2021) produced the meta-analysis intended for use in health impact assessment: each 1 dB of aircraft noise at school associates with a change of −0.007 (95% CI −0.012 to −0.001) in reading score and approximately 4% higher odds of scoring below or well below average; hyperactivity +0.017 (0.007–0.028) per dB [documented].

Adults, and the office

Here the harm occurs at levels a sound meter calls quiet — far below any hearing-protection limit, and often below environmental guidelines too.

Kim and de Dear, analysing the Berkeley Center for the Built Environment occupant database in 2013, found that over 59% of open-plan occupants were dissatisfied with sound privacy — the largest single source of dissatisfaction in that layout — with partitions largely ineffectual [documented]. An earlier draft of this study reported this as "the single largest source of dissatisfaction" full stop; the finding is specifically among open-plan occupants, and dissatisfaction is not a health outcome. This is cross-sectional self-report.

The mechanism is the irrelevant speech effect: intelligible background speech disrupts tasks requiring serial rehearsal and short-term memory, and the disruption tracks primarily with speech intelligibility — measured as the Speech Transmission Index — rather than with sound level [documented]. An earlier draft said "with intelligibility, not level," which overstates a real finding into an absolute. Primarily, not exclusively.

This is why ISO 3382-3 specifies a "distraction distance," and why the design practice of sound masking exists at all: raising a steady, non-informational background specifically to destroy intelligibility [documented as mechanism; contested as to field effect magnitude].

Specific level figures often quoted for the office literature — around 39 dB LAeq for a low condition and 51 dB for a high one, or a surveyed office range of 29–45 dB LAeq — could not be verified against the cited sources and have been removed [unverifiable]. The paragraph's point survives without them.

The opposite: what quiet and natural sound do

Buxton and colleagues, in PNAS in 2021, reviewed 36 publications and meta-analysed 18. Natural sound exposure was associated with decreased stress and annoyance, g = −0.60 (95% CI −0.97 to −0.23) [documented], and with improved health and positive affect, g = 1.63 (95% CI 0.09 to 3.16) [contested].

Look at that second confidence interval. It runs from 0.09 — trivially small — to 3.16, which would be one of the largest effect sizes in behavioural science. Across 18 studies, an interval that wide is what small-study and publication bias look like. The direction is supported; the magnitude is uninformative. We are flagging this ourselves because it is the sort of number that gets turned into a headline.

Alvarsson, Wiens and Nilsson (2010, n = 40) reported faster skin-conductance recovery after a stressor during nature sound than during noise, with no effect on heart-rate variability [contested — small, the effect was described as a tendency, and the publisher blocked re-verification of the exact wording].

8. Hearing itself: the one exposure people choose

This section is kept deliberately separate, because merging it with everything above is a common trick that makes the evidence look larger than it is.

Environmental noise is involuntary and its documented endpoints are cardiometabolic and cognitive. Personal audio is voluntary and its endpoint is the hearing organ. Different exposures, different mechanisms, different levers. Stacking them is not argument, it is inflation.

Dillard and colleagues, in BMJ Global Health in 2022, pooled 33 studies covering 19,046 participants aged 12 to 34. Estimated prevalence of unsafe listening: 23.81% (95% CI 18.99–29.42) for personal listening devices and 48.20% for loud entertainment venues, placing an estimated 0.67 to 1.35 billion young people at potential risk of hearing loss [documented, with wide bounds and high between-study heterogeneity acknowledged by the authors].

Two honest qualifications. "At risk" is not "will lose hearing." And the range is 0.67 to 1.35 billion — a factor of two. An earlier draft of this study quoted only the upper bound. Quoting only the top of a range is the single most common way an honest number becomes a dishonest sentence.

The one randomised trial worth knowing here: Ramakers and colleagues, JAMA Otolaryngology, 2016, randomised 51 attendees at an Amsterdam music festival to earplugs or none. Temporary threshold shift at 3–4 kHz occurred in 4 of 50 ears (8%) protected versus 22 of 52 (42%) unprotected, relative risk 5.3 (2.0–14.3); new tinnitus 3 of 25 (12%) versus 10 of 25 (40%) [documented].

That is a striking result and it has real limits, which we state before anyone else does: n = 51, one festival, single-blind, and temporary threshold shift is a recognised surrogate for cumulative damage — not a measurement of permanent hearing loss [documented limit].

The reason this section exists at all, given that it is a different exposure, is that it is where the personal decibel intuition can be calibrated. The NIOSH occupational limit is 85 dB(A) over eight hours with a 3 dB exchange rate: 88 dB for four hours, 91 dB for two, 94 dB for one [documented, calculated from the exchange rate]. Amplified music venues routinely exceed all of those. Environmental guidelines, by contrast, sit thirty to forty decibels lower, because they are protecting something other than the ear.

9. Where music genuinely heals

Everything so far has been subtractive: what happens when sound is done to a body. This section is the additive half — what happens when sound is offered deliberately — and it is included because a study that only ever says "no" to sound-as-medicine would be as unbalanced as one that says yes to everything.

The evidence here is real, replicated, and unevenly strong. It is largest for short-term procedural pain and anxiety, strong for rhythmic gait rehabilitation after stroke, modest for depression as an add-on, and null for several outcomes people confidently assume are its best.

A distinction the literature enforces

Music therapy means delivered by a trained music therapist, relational and tailored. Music medicine means pre-recorded listening offered by medical staff. Bradt and colleagues, in their 2021 Cochrane review of music interventions in cancer care (81 studies, 5,576 participants), state that for several outcomes therapist-delivered music therapy "led to consistent results across studies and this was not the case for music medicine interventions" [documented]. Popular summaries routinely merge the two [contested — asserted from reading, not measured; an earlier draft of this study called it "the single most common error in popular summaries," which is an editorial claim with no bibliometric source behind it].

Where the evidence is strongest

Perioperative. Hole and colleagues, The Lancet, 2015 — 73 randomised trials, 6,902 patients [documented]:

Timing — before, during or after surgery — made little difference, and effects persisted under general anaesthesia at a smaller magnitude [documented]. That last finding is the single result most in tension with a pure attention-and-expectancy account of how music works, and it remains [contested]. Note also the null: length of stay unchanged. Real analgesic effect, no effect on the outcome hospitals actually budget for.

Pooling standardised mean differences across heterogeneous pain scales from unblinded self-report is the standard attack on this review, and we could not retrieve its heterogeneity statistics or funnel-plot results to pre-empt it [unverifiable]. So we are not quoting any I² for it, in either direction.

Paediatric procedures. A 2022 meta-analysis in Journal of Clinical Medicine (38 randomised trials, 5,601 participants) found significant pain reduction in newborns and in infants and children, with reduced heart and respiratory rate and increased oxygen saturation [documented]. It pools newborns with older children and is overwhelmingly music medicine rather than music therapy.

Cancer care. Bradt et al., Cochrane 2021, with the certainty ratings kept adjacent to every number, which is the only honest way to present them [documented]:

Real numbers, weak confidence. Both facts have to travel together or the table is a lie.

Stroke gait — rhythmic auditory stimulation. This is the front with the clearest mechanism, and two meta-analyses that do not closely agree [documented]:

An earlier draft of this study presented these as a single range — "velocity 0.73–0.98, stride 0.58–0.76, cadence 0.75–0.84" — which matches neither source. Fabricated ranges are exactly the failure this house exists to refuse, and it is logged. The dose in Ghai and Ghai is 20 to 45 minutes, 3 to 5 times per week [documented].

Note what the two reviews imply together: the smaller, older review reports the larger effects. That is the classic signature of small-study bias [contested].

Critical care. Bradt & Dileo, Cochrane 2014 (14 trials, 805 participants): anxiety SMD −1.11 (−1.75 to −0.47); respiratory rate and systolic blood pressure consistently reduced; heart rate mixed; oxygen saturation unimproved; sedative and analgesic intake reduced in one large trial with trends in two others [documented]. Most trials at high risk of bias from absent blinding, per the review itself.

Depression. Aalbers et al., Cochrane 2017 (9 studies, 421 participants): music therapy added to treatment-as-usual beat treatment-as-usual alone — clinician-rated SMD −0.98 (−1.69 to −0.27; 4 studies, n = 219), patient-rated −0.85 (−1.37 to −0.34; 4 studies, n = 142), moderate quality. Anxiety and functioning also favoured. Quality of life null (0.32, −0.17 to 0.80, low quality). No long-term data at all [documented].

Where it fails, stated as plainly as where it works

Dementia agitation. van der Steen and colleagues, Cochrane 2025 (CD003477.pub5), report moderate certainty that music-based interventions likely do NOT improve agitation or aggression — SMD −0.05 (−0.27 to 0.17; 11 studies, 503 participants) [documented]. Depressive symptoms probably improve slightly, −0.23 (−0.42 to −0.04; 9 studies, 441), moderate certainty [documented]. Overall behavioural problems −0.31 (−0.60 to −0.02; 10 studies, 385), low certainty [documented]. Emotional wellbeing, anxiety, social behaviour and cognition: no improvement, low to very low certainty. Beyond four weeks after treatment, probably no advantage over usual care [documented].

This is a null with moderate certainty, which is rarer and more informative than most positive findings in this field.

Care-home depression at scale — MIDDEL. Eighty-six care-home units across six countries. At the primary endpoint — MADRS at 6 months, 751 residents with data — neither group music therapy nor recreational choir singing beat standard care [documented]. Exploratory per-protocol country analyses favoured choir singing in Australia, Norway and Türkiye, went the wrong way in the United Kingdom, and were null in Germany and the Netherlands [documented]. A 2022 Australian cohort of the same trial had supported choir singing; the multinational analysis did not confirm it.

An earlier draft of this study said "country was the strongest predictor of effect." That is a ranked statement and no source supports it as a ranking [unverifiable]. Per-protocol country splits are exploratory, not a predictor hierarchy.

Preterm infants — LongSTEP. Parent-led infant-directed singing had no effect on mother-infant bonding: NICU arm 0.55 (−2.20 to 3.30), post-discharge 1.02 (−1.72 to 3.76) at 6 months with 196 dyads; still null at 12 months with 181; secondary outcomes null. Safe and well accepted [documented].

Blood pressure — and this is the correction that would have discredited this study. The figure that circulates is a systolic reduction of about 10 mmHg, from 144 down to 134. That is a within-group, before-and-after change from a 2016 BMC Cardiovascular Disorders review of 10 randomised trials, which explicitly declined to establish a cause-effect relationship [documented]. The pooled between-group effect — the one that actually means anything — is systolic −2.63 mmHg (−3.91 to −1.34), diastolic −1.11 (−1.69 to −0.53), heart rate −3.42 bpm (−5.03 to −1.81) [documented, Loomba et al.]. An earlier draft of this study reported the within-group figure as though it were the effect, overstating by roughly fourfold. That one line would have been enough for a hostile reader to dismiss the entire document, correctly.

Aphasia — melodic intonation therapy. Koshimori et al., 2025, Annals of the New York Academy of Sciences: 10 randomised trials, gains in repetition and naming, moderate-to-high risk of bias, and functional communication examined in just two trials [documented]. An earlier draft added "phrase length" to the list of improved outcomes. It is not in the source. Removed.

Mechanisms, and the bias that cannot be removed

Dopamine: Salimpoor and colleagues, Nature Neuroscience 2011, demonstrated striatal dopamine release during pleasurable music — caudate during anticipation, nucleus accumbens at the moment of peak pleasure [documented]. Endogenous opioids: Mallik and colleagues (2017) found naltrexone attenuated both positive and negative musical emotion [documented], but two later pharmacological studies did not replicate the subjective-pleasure effect while still shifting physiological arousal markers [contested]; a 2024 μ-opioid PET study reporting receptor activation by pleasurable music was at preprint status when checked and is treated as [contested].

And the fact that governs everything above: blinding is impossible in a music trial. [documented] A participant always knows whether music is playing. The direction of expectancy bias is therefore upward, by an unknown and unmeasured amount.

An earlier draft of this study wrote that "every effect size above is inflated by expectancy [documented]." Two errors in eight words: the inflation is an inference, not a documented quantity, and "every" is wrong. Objectively measured outcomes — gait velocity, respiratory rate, analgesia consumption — and trials with active control conditions are considerably less exposed to it [contested — inference]. Corrected.

What must be refused

That music heals at a particular frequency, or that particular frequencies address particular organs [myth]. That music therapy substitutes for analgesia, antidepressants or rehabilitation [myth] — every positive finding above is adjunctive, added on top of standard care. That the dementia-agitation null makes music useless in dementia [myth] — the depression and behavioural findings stand on their own. That music lowers blood pressure by about 10 mmHg [myth].

Sound healing with singing bowls: Stanhope and Weinstein, Complementary Therapies in Medicine, 2020, found four peer-reviewed studies and stated they cannot recommend singing bowl therapies at this stage [documented]. The methodological quality range often quoted alongside this is unverified [unverifiable]. "Cannot recommend at this stage" is not the same as "does not work"; it means the studies to decide have not been done.

10. Entrainment, honestly

Binaural beats are the most respectable member of the sound-wellness family, and they deserve a careful hearing rather than a dismissal — partly because the underlying perceptual phenomenon is genuinely real and strange.

The physics is real. The brainwave claim is a separate claim.

Present 400 Hz to one ear and 410 Hz to the other, and no 10 Hz sound exists in either ear canal. The beat is constructed centrally [documented — Oster, Scientific American, 1973; note this is a popular-science article, not a peer-reviewed paper]. The superior olivary complex is named in the literature as the main neuroanatomical structure involved [documented as a literature attribution, not as a demonstration].

Perception is bounded in two ways: carrier tones must be at most about 1000 Hz [documented — Licklider et al., via Ingendoh 2023], and beyond a difference of about 30 Hz the two tones are simply heard as two tones [documented — Perrott & Nelson, via Ingendoh 2023]. An earlier draft of this study gave the carrier ceiling as "roughly below 1000–1500 Hz" and added a claim about best discrimination around 10–15 Hz. The 1500 has no source; the discrimination claim is not in the cited review at all. Both removed.

Oster's own stated interest was diagnostic — beats as a probe of binaural hearing function [documented]. The entrainment framing was attached later [contested]. We soften this from an earlier draft which asserted Oster "never proposed brainwave entrainment"; that is an argument from absence, in a magazine article we could not open in full.

Isochronic tones are a different physical object

An isochronic tone is a real amplitude modulation in the air — the sound genuinely pulses. A binaural beat does not. The closest measured comparison is monaural or acoustic beats: a 40 Hz binaural-beat auditory steady-state response was evoked at 400 Hz carriers but became undetectable beyond 3 kHz, and its amplitude was smaller than the acoustic-beat response [documented — Schwarz & Taylor, Clinical Neurophysiology, 2005]. Extending that to isochronic tones is extrapolation: a monaural beat and a gated tone are not the same stimulus [contested].

A claim circulates that the research literature is split roughly 15 studies on binaural beats to 2 on isochronic. The review it is attributed to contains no mention of isochronic or monaural tones at all [documented absence], and the companion citation names no author or title. The split may or may not be real; it is unverifiable and the specific numbers have been deleted.

Does EEG actually entrain? Mostly no.

Ingendoh, Posny and Heine's 2023 PLOS ONE review found, across 14 studies: 5 supported entrainment, 8 contradicted it, 1 was mixed [documented]. Sample sizes ran from 4 to 47 participants. In the review's own words, "twelve out of the 16 studies did not implement a control group design" — we quote the paper rather than smooth over the denominator mismatch, which is the paper's, not ours. Heterogeneity in how EEG was operationalised made meta-analysis impossible [documented].

A dedicated attempt to detect the effect found no EEG band change and no arousal change — but with n = 14, and it found nothing for acoustic beats either, which cuts against the isochronic story as much as the binaural one [documented — Frontiers in Human Neuroscience, 2017].

The mechanism most often invoked is the one least supported. That said, the negative studies are themselves too small to settle it [contested].

Behavioural and affective outcomes: modest and fragile

Garcia-Argibay, Santed and Reales, Psychological Research, 2019: pooled across cognition, anxiety and pain, g = 0.45 from 22 studies and 35 effect sizes [documented]. Masking with white or pink noise was unnecessary; exposure before, or before and during, the task worked; longer exposure helped.

Anxiety specifically, with theta or delta beats: g = 0.69 — from 5 effect sizes across 4 studies, N = 159 [documented value, contested weight]. That is a hypothesis, not a finding.

Basu and Banerjee, Psychological Research 87 (2023): memory and attention g = 0.40, 15 studies, 31 effect sizes, with conflicting theta and beta results [documented]. An earlier draft of this study marked this count "unverifiable" when the number is plainly available. Leaving "unverifiable" where a verifiable figure exists is its own kind of failure, and it is logged.

Clinical settings: largest effects, largest warnings

Xiong and colleagues, Complementary Therapies in Medicine, 2025: perioperative anxiety, 14 trials, n = 1,047, SMD −1.38, with I² = 91.6%; pain, 5 trials, n = 433, SMD −0.61, I² = 75.8% [documented].

That heterogeneity figure is not a footnote. With I² of 91.6% and no funnel plot or Egger test retrieved, the pooled value cannot be read as a treatment effect, and small-study bias is at least as likely an explanation as genuine clinical variation [contested]. An earlier draft of this study quoted the −1.38 without the I². That is the kind of omission that turns a real result into a misleading one.

Dental anxiety: Shukla et al., International Journal of Clinical Pediatric Dentistry, 2025 — between-group MD −1.50 on the visual analogue scale, I² = 0%, which is clean. But the same paper's between-group pain result is null: MD −3.30 (−37.67 to 31.06), I² = 94%; risk of bias was high in allocation concealment and detection; and GRADE certainty is LOW, with the authors stating results "should be extrapolated with caution" [documented]. Six studies meta-analysed from nine reviewed. Reporting the anxiety figure without the null pain figure would be a cherry-pick.

Sleep: subjective gains, no objective validation

A single-blind randomised trial in Taiwan (n = 64, long-term care, 14 days) of binaural beat music reported better questionnaire-rated sleep, heart-rate-variability shifts and lower depression scores [documented — Geriatrics & Gerontology International 2024;24(3):297–304; an earlier draft attributed this to the wrong journal]. No polysomnography. Whether binaural beats change sleep architecture rather than sleep report is unverifiable.

The adjacent line doing better — with its own caveats

40 Hz gamma sensory stimulation (light and sound together, not binaural beats) has produced the most interesting recent result in this space: corpus callosum area preserved in the active group versus atrophy in sham at 6 months, 50 participants randomised 2:1, p < 0.02 [documented — Da et al. 2024].

An earlier draft of this study called that "a 6-month RCT." It is a retrospective, non-pre-specified MRI analysis of the OVERTURE Phase 2 trial (NCT03556280), funded by Cognito Therapeutics, with company-salaried authors, on a device the sponsor sells [documented]. Saying so is not an accusation; it is disclosure that the reader is entitled to. The two-year extension enrolled five patients, open-label, with external database controls, and plasma pTau217 available for two [documented]. No published Phase 3 result was retrieved [unverifiable].

The contrast with binaural beats is that here there is a stated mechanism and a biomarker endpoint — not that the evidence is yet strong [contested as interpretation].

Verdict

Binaural beats are a genuine perceptual phenomenon with a named anatomical substrate [documented]. The claim that they entrain cortical oscillations is contradicted more often than supported, in a literature too small and too heterogeneous to settle either way [documented]. Small-to-moderate effects on anxiety and attention recur across independent meta-analyses [documented].

One rhetorical line an earlier draft of this study leaned on has to go. It said "almost no trial isolates the beat from the music carrying it." Active-audio comparators exist and still favour the beat: Xiong 2025 pooled binaural against non-binaural audio and found anxiety SMD −1.01 across 8 trials with n = 598, and pain −0.52 across 3 trials with n = 265 [documented]. The correct, weaker, honest claim: active-audio comparators exist and still favour the beat, but none retrieved were blinded to expectancy, so the effect remains fully compatible with expectancy and attentional distraction.

The structural point. Entrainment is the additive proposition — add a tone, gain a state — and its evidence is mixed at best. The subtractive proposition — remove the noise, recover the sleep — is far better evidenced, at vastly larger scale, and costs nothing. That asymmetry is the shape of this entire study.

11. The Schumann resonance and EMF: two questions, one conflation

This section exists because the 432 narrative almost always arrives fused with a second one: that the Earth has a natural frequency, that human beings are tuned to it, that modern electromagnetic environments have severed the connection, and that this is part of the same harm.

There are two entirely separate subjects here. The Schumann resonance is geophysics. Phone, wifi and power-line electromagnetic fields are a public-health question. They share the word "electromagnetic" and almost nothing else, and nearly every claim that fuses them does so on the strength of that shared word alone.

(a) The Schumann resonance — real physics, badly reported

Predicted by Winfried Otto Schumann in 1952; reliably measured by Balser and Wagner between 1960 and 1963 [documented]. The gap between the Earth's surface and the ionosphere behaves as a resonant cavity, excited by global lightning activity [documented]. The fundamental mode sits near 7.83 Hz [documented], with higher modes conventionally given as 14.3, 20.8, 27.3 and 33.8 Hz — noting that published values vary by measurement campaign, and that one standard reference lists 14.1, 20.3, 26.3 and 32.5 Hz [documented, source-dependent]. An earlier draft of this study attributed one set of harmonics to a source that gives the other. Corrected.

The physics error we made, and it is the one a physicist checks in thirty seconds. An earlier draft said the theoretical ideal-cavity value is about 7.5 Hz, and that the observed 7.83 Hz sits slightly above it. Schumann's ideal-cavity formula is f_n = (c/2πa)√(n(n+1)), which gives approximately 10.6 Hz for the fundamental [calculated]. The 7.49 Hz figure is the prefactor c/2πa, not the fundamental. The observed 7.83 Hz is therefore below the ideal value, not above it — the discrepancy points the opposite way from what we wrote — and the reason is that the real ionosphere is a lossy, moving boundary rather than a perfect conductor [documented].

Amplitude settles most of the argument. The Schumann resonance magnetic signal is on the order of 1 picotesla — described in the standard reference as many orders of magnitude smaller than the Earth's static magnetic field of roughly 30 to 50 microteslas [documented]. That is approximately seven orders of magnitude [calculated]. You are standing, continuously, in a static field ten million times stronger than the signal in question.

"The Earth's frequency is rising to 40 Hz." [myth] The fundamental has stayed near 7.8 Hz, drifting roughly between 7.5 and 8.1 Hz with ionospheric conditions [documented; an earlier draft gave 7.5–8.3, slightly too wide]. The ~33.8 Hz band people point at is the fifth mode, the fourth overtone, present in 1960s measurements [documented; an earlier draft called it the "fourth harmonic," which normally means 4× the fundamental]. The claim appears to originate in misreading public spectrograms in which colour encodes amplitude rather than frequency — but the source we cited for that explanation returns a 404, so it is offered as a plausible inference, not an established fact [unverifiable as cited].

Biological coupling. Cherry (2002, Natural Hazards) proposed the Schumann resonance as a mechanism linking geomagnetic activity to health [contested — a hypothesis paper, not a demonstration]. The much-repeated link between 7.83 Hz and the brain's alpha rhythm rests on adjacency, not overlap: alpha is conventionally 8 to 13 Hz, and 7.83 sits just below it [documented]. An earlier draft of this study said the overlap "is arithmetically real." It is not. It is adjacency, and a paragraph that claims an overlap while stating two non-overlapping numbers refutes itself.

A 2025 study in Cureus, titled for the Schumann frequency and cited in wellness contexts as evidence of coupling, applied a field of 1,300 Gauss = 0.13 tesla — about eleven orders of magnitude above the natural picotesla signal — to six men, with no sham control [documented from the paper itself]. It is not evidence about the Schumann resonance. It is a useful illustration of how the label gets attached to fields that are nothing like the phenomenon.

The believer's strongest reply, which we are going to state rather than dodge: resonance is frequency-specific, not amplitude-driven, so comparing picoteslas to microteslas is beside the point. That objection is not answered by the amplitude argument alone. The honest response is that a proposed resonance still has to exceed the biological noise floor to be detectable — but we could not source that quantitatively, so it stands as the open question rather than as a rebuttal.

The commercial claim that buildings or cars shield you from the Schumann resonance, creating a deficiency that a purchasable device restores, is not supported by any measurement study we found [unverified — and note that "we found no study" is weaker than "no study exists"; we are stating the weaker thing].

(b) Non-ionising EMF — settled, open, and overrun

Settled. ICNIRP's 2020 guidelines (100 kHz to 300 GHz) protect against the established adverse effects: whole-body heating, using a 4 W/kg whole-body-average specific absorption rate as the adverse-effect threshold averaged over 30 minutes, with a reduction factor giving 0.08 W/kg for the general public — plus the ICNIRP 2010 nerve-stimulation restrictions for 100 kHz to 10 MHz [documented]. An earlier draft called heating "the one established adverse mechanism"; nerve stimulation is also in there, and the sentence has been corrected. Whole-body restrictions now span the full range rather than stopping at 10 GHz, and — the change ICNIRP itself names as the significant one — the local-exposure transition frequency moved from 10 GHz to 6 GHz [documented].

Real-world magnitude. In one 2020 comparative measurement study of Wi-Fi 4 and Wi-Fi 5 devices, exposures reached 0.0001% (802.11ac) to 0.0064% (802.11n) of the ICNIRP reference level for local exposures under six minutes [documented]. That is one study, not a survey, and it is doing a lot of work in public communication that a single paper cannot carry.

People who suffer. Rubin and colleagues' 2010 update in Bioelectromagnetics — 46 blind or double-blind provocation studies, 1,175 participants — found no robust evidence that people reporting electromagnetic hypersensitivity can detect exposure; symptoms track believed exposure, a nocebo pattern [documented]. An earlier draft of this study cited the 2010 numbers against the 2005 paper's identifier (31 experiments, 725 participants); corrected.

Rule 5 applies here as hard as anywhere in this study. The symptoms are real and sometimes disabling. Only the attributed cause is unsupported. A person whose life has narrowed around an explanation that turns out not to hold has still lost the life, and has done nothing to be ashamed of.

Genuinely open. IARC classified radiofrequency fields as Group 2B in 2011, on limited evidence for glioma in heavy users [documented]. WHO's commissioned systematic reviews found moderate certainty that mobile phone use does not increase glioma, meningioma, acoustic neuroma, pituitary, salivary gland or paediatric brain tumours — Karipidis et al. 2024, Environment International 191:108983, covering 63 publications reporting 119 studies from 1994 to 2022 — with lower certainty for cordless phones and lower still for occupational and childhood environmental exposure [documented]. Less-researched outcomes, including lymphatic and haematopoietic cancers, thyroid, and oral cavity and pharynx, are covered in Part II (2025; 26 articles, 143 studies) [documented]. An earlier draft of this study fused both papers under one citation and invented a study count of 74. Corrected.

ICBE-EMF (Melnick, Moskowitz, Hardell and colleagues, Environmental Health, 2025) argues that these reviews provide no assurance of safety [contested]. An earlier draft listed ICBE-EMF and "a 2025 critique" as two sources. They are the same body and the same paper. The dissent is real; it is counted once.

The US National Toxicology Program (2018) and the Ramazzini Institute both found heart schwannomas in male rats. ICNIRP's 2020 Note judged the dose pattern unreliable for revising limits; the critics read the same data as one tumour type appearing at doses differing by a factor of about sixty [contested — both readings attributed rather than one presented as neutral fact].

Decisively for anyone who wants to know whether this file is closed: the IARC Monographs advisory group, writing in The Lancet Oncology in April 2024, named radiofrequency fields a high priority for re-evaluation in 2025–2029, citing new human and animal cancer evidence [documented]. Anyone declaring this settled, in either direction, is ahead of the evidence.

Open, and older. Extremely low frequency magnetic fields are Group 2B (IARC 2002). Ahlbom and colleagues (2000, British Journal of Cancer; 3,203 cases, 10,338 controls) found a summary relative risk of 2.00 (1.27–3.13) for childhood leukaemia at 0.4 µT and above — in the 0.8% of children exposed above that level [documented]. An earlier draft of this study said 1–4% of children; that figure is not in the cited paper. No mechanism is established and selection bias remains a live explanation [documented].

And here an earlier draft stopped, twenty-four years short — which cost us the two strongest pieces of evidence for our own verdict. It ended on SCENIHR (2015), a document that has since been superseded. Both of the following were missing, and both point the same way.

The newest pooled analysis is flat null. Amoon, Swanson, Magnani, Johansen and Kheifets, Environmental Research 204 (2022), PMID 34481821: four recent studies, individual-level data, 24,994 cases and 30,769 controls — nearly eight times Ahlbom's case count — found no increased risk, with an odds ratio of 1.01 at ≥0.4 µT compared with <0.1 µT [documented]. That is the same exposure contrast on which Ahlbom found a doubling, in the largest and most recent individual-level pooling available, and it comes out at unity.

And the European scientific committee has downgraded the evidence in writing. SCHEER — the Commission's Scientific Committee on Health, Environmental and Emerging Risks — adopted its updated opinion on EMF between 1 Hz and 100 kHz on 17 May 2024. Its finding, quoted exactly: "overall, there is weak evidence concerning the association of ELF-MF exposure with childhood leukaemia." It reaches that by three lines of evidence — "weak to moderate weight of evidence from epidemiological studies (the primary line of evidence)," weak evidence from animal models it judges largely inappropriate to the question, and weak evidence from interaction mechanisms [documented — SCHEER, final opinion, adopted 17 May 2024]. The Commission's own summary of the position is that the epidemiological association exists but that "no mechanisms have been identified and no support from experimental studies could explain these findings, which, together with shortcomings of the epidemiological studies prevent a causal interpretation."

Why we are recording this as an error against ourselves rather than an update. Our verdict was "real, small, unresolved," and both omitted items strengthen it. Leaving them out was not spin — it made our own conclusion weaker than the evidence allowed, which is the failure mode a study running on caution should expect to have. But it is still a failure: we cited a 2015 assessment as current when a 2024 committee opinion and a 2022 pooled analysis of twenty-five thousand cases both existed and both said what we were already saying. The verdict now reads: real, small, unexplained — and shrinking. An association that halves in strength each time the sample grows is behaving the way a confounded association behaves, and nobody is entitled to say more than that yet. And it still has nothing whatever to do with 7.83 Hz.

12. What the Earth itself sounds like

The section above took one planetary signal apart. This one takes the rest of the family, because the Schumann resonance almost never travels alone — it arrives bundled with three other phenomena, all of them described with the same handful of words, and the bundling is where the reasoning goes.

Start with the word itself. "Sounds like" is already wrong, and it is worth losing before anything else. Four continuous planetary signals are routinely blended into a single "Earth frequency." They are not one thing, and they are not one kind of thing. Two of them are electromagnetic — the Schumann resonances, and geomagnetic pulsations. Two are mechanical: elastic motion of the ground itself — microseisms, and the Earth's hum. None of the four is audible, and for two different reasons. The mechanical pair are far too slow and far too faint. The electromagnetic pair are not pressure waves in air at all, whatever their number in hertz — a 20 Hz magnetic oscillation is no more a sound than a radio station is [documented]. "Hum" is a borrowed word here. It is doing metaphorical work while being read as literal, and almost every claim in this territory depends on that slippage going unnoticed.

The Schumann resonances, and one further thing about the derivation

Section 11 corrected our own physics error: the ideal-cavity fundamental is about 10.6 Hz, not 7.5, and the observed 7.83 Hz therefore sits below the ideal value rather than above it. One consequence of that correction deserves stating outright, because it dissolves the most persuasive argument in the whole area.

The familiar move runs: take the speed of light, divide by the Earth's circumference, and you get about 7.5 Hz — look how close that is to the measured 7.83. It sounds like derivation. It is not. That calculation produces the prefactor c/2πa, and the fundamental is that prefactor times √2 [calculated]. Dropping the √2 lands you near the right number by accident, and then the small remaining gap gets credited to ionospheric loss — an effect whose actual sign runs the other way, and whose actual size is about a quarter of the ideal value, not a rounding error [documented]. The agreement is a coincidence wearing the clothes of a derivation. A study about numerical adjacency should be the first to say so about a calculation on its own side of the argument.

Two further details, both of which undercut the idea of a fixed planetary pitch. The source is lightning, not the planet: roughly 2,000 thunderstorms active at any moment, about 50 flashes per second on the standard reference (satellite climatologies give ~44–46 s⁻¹) [documented]. And the peak moves. Zhou, Yu, Cao and Qiao (JASTP 98:86–96, 2013) found the frequencies of all four modes at maximum in boreal winter and minimum in summer, with amplitudes running opposite [documented]. During a solar proton event on 14 July 2000, the first mode's frequency rose during the preceding X-ray burst and fell during the proton event itself (De et al., JASTP 72, 2010) — first mode only, and the magnitude of that shift is unverifiable from the sources we reached [documented as to direction]. So 7.83 Hz is the long-run average of a wandering peak, not a constant. It is a poor thing to lock onto.

Microseism — the ground, moving, all the time

This one is genuinely mechanical and genuinely continuous, and almost nobody in the wellness literature mentions it, which is telling in itself.

Ocean waves shake the solid Earth. Primary microseisms occur at ocean-wave frequencies — periods of about 4 to 30 seconds, peaking near 16 seconds — where swell interacts with sloping seafloor. Secondary microseisms occur at double that frequency, strongest around 0.1 to 0.2 Hz, produced when opposed wave trains create standing pressure oscillations that reach the seabed: the Longuet-Higgins (1950) and Hasselmann (1963) mechanism, dominantly Rayleigh waves with a significant Love-wave component [documented]. Ground amplitude "does not generally exceed 10 micrometres" [documented].

Ten micrometres at 0.1 Hz is an acceleration of a = (2πf)²d ≈ 3.9 × 10⁻⁶ m/s² [calculated]. Now notice what cannot honestly be said next, because our own draft said it. ISO 2631-1 defines whole-body vibration perception over 0.5 to 80 Hz, and treats 0.1 to 0.5 Hz as motion sickness instead — a different quantity with a different weighting [documented]. The secondary microseism peak therefore sits below the band in which a perception threshold is defined at all. So the correct statement is not "several thousand times below the threshold." It is: orders of magnitude below any threshold defined at neighbouring frequencies, and at its own frequency the standard declines to define one. It is instrument-detectable, not body-detectable. The conclusion survives; the shortcut to it did not.

The Earth's hum — the planet ringing on quiet days

Discovered in 1998, independently by several groups — Nawa and colleagues on a superconducting gravimeter at Syowa, Antarctica; Suda and colleagues; Kobayashi and Nishida; Tanimoto and colleagues — the finding is that the planet's normal modes ring continuously, even on days with no earthquakes [documented].

Rhie and Romanowicz (Nature 431:552–556, 2004) identified free-oscillation peaks at 2 to 7 mHz on quiet days and traced the excitation to atmosphere–ocean–seafloor coupling, at "a level of daily excitation … equivalent to that of magnitude 5.75 to 6.0 earthquakes" — far more than the sum of small quakes could supply [documented]. Nishida's review (Annu. Rev. Earth Planet. Sci. 41:719–740, 2013) gives the excited band as 2 to 20 mHz — 2 to 7 mHz was Rhie and Romanowicz's analysis window, not the extent of the phenomenon — and reports ocean infragravity waves coupling to sloping seafloor as the leading source, while adding that surface pressure sources, plausibly atmospheric turbulence, remain significant below 5 mHz [documented]. That is worth flagging: the excitation is well constrained, and it is not closed. Anyone presenting it as settled is presenting a review's headline without its qualification.

Amplitude, per mode, is "on the order of 0.5 nGal (10⁻¹¹ m s⁻²)" — quoted as Nishida writes it, noting that 0.5 nGal is strictly 5 × 10⁻¹² m s⁻²; the rounding is the source's [documented]. Detecting it needs a superconducting gravimeter or a very quiet broadband seismometer. Two millihertz is a period of about eight minutes. Nothing about it is available to a human sense.

Geomagnetic pulsations — electromagnetic, and not the Earth's own

Ultra-low-frequency magnetic oscillations, classified Pc1–Pc5 (continuous) and Pi1–Pi2 (irregular) after Jacobs and colleagues in 1964 [documented]. The band runs from about 1.7 mHz (Pc5) to 5 Hz (Pc1) — not, as commonly written including in our own first draft, up to 1 Hz [documented]. Pc1 amplitudes are typically 1 to 5 nT; Pc5 can reach a few hundred nT, maximal in the auroral oval [documented].

The claim that storm-time Pc5 exceeds quiet-time levels by two orders of magnitude is plausible and we could not stand it up: the paper usually cited for it — Yagova, Pilipenko, Sakharov and colleagues, Earth Planets Space 73:88 (2021) — is about geomagnetically induced currents in a Kola Peninsula power line, not about amplitude ratios [documented as to what the paper is]. Tag: unverifiable.

And the attribution matters. The driver is the solar wind coupling to the magnetosphere. Earth supplies the resonant cavity; the Sun supplies the hammer [documented]. This is not the planet's voice in any sense the devotional reading wants.

So what could "retuning the Earth," or being retuned by it, physically mean?

The phrase has to cash out as one of three things. (a) An oscillator in the body phase-locks to the signal. (b) Tissue resonantly absorbs energy at that frequency. (c) Metaphor. Only (c) is currently defensible, and (c) is not a small thing — but it should be named as what it is.

For (b): a 1 picotesla field is about seven orders of magnitude below the Earth's own static field, and vastly below any low-frequency exposure reference level. One caution about how that last comparison is usually made, including in our own drafting. ICNIRP's 200 µT figure is the general-public reference level at 50/60 Hz, not at 7.83 Hz [documented]. The low-frequency reference levels are frequency-dependent, and the value near 8 Hz is higher still — we could not render the ICNIRP table to give the exact number, so it stands as unverifiable. Comparing a 7.83 Hz field against the 50 Hz limit is a category slip, and we are naming our own. It rescues nothing: every reference level in that band is many orders of magnitude above a picotesla, and no tissue structure with a resonance and a quality factor sufficient to accumulate energy from such a field has been demonstrated [documented].

For (a): the alpha-band coincidence was dealt with in section 11 — adjacency, not overlap. Add the moving-target problem above and the case gets weaker rather than stronger.

And there is a fourth answer, on the other half of the phrase — "retuning the Earth" in the ordinary sense of retuning all the world's music. It was missing from every draft of this study, and it is the cleanest physical objection in the file. Shift music down by 31.77 cents and every partial in it moves. The rooms do not. Room modes are set by the dimensions of the space and the speed of sound in it. Loudspeaker and cabinet resonances are set by enclosure geometry and driver parameters. The ear canal's own quarter-wave resonance, near 2.5–3 kHz, is set by the length of a canal you were born with. None of the three moves by a single cent when the music is retuned [deduction from standard acoustics — each mechanism is textbook; the application here is ours].

What that means is exact. A global retune would not deliver one acoustic result to everybody; it would change the relationship between the music and a different set of fixed resonances in every room and every skull. There is no configuration in which "the world at 432" is a single physical state. This costs nothing to concede, disputes none of the historical or arithmetic claims, and survives granting all of them — which is why it is a stronger argument than any of the numerology, and why its absence from our drafts is recorded in the corrections log.

And there is a plainer objection that costs nothing to state, and that we mark as a deduction rather than a measurement, because that is what it is. You are already inside all four, permanently, and you cannot leave. The Schumann cavity has no outside for a person on this planet. Microseisms shake the ground under every building. So "realignment" cannot mean restoring an exposure, because the exposure is unconditional and always was. The framing appears to import a grammar of separation-and-return from spiritual practice into a domain where the physical premise — that one could be out of it — is simply false [contested, interpretive].

On devices: a commercial "Schumann generator" emits a locally produced ELF field. Vendors generally publish no field-strength specification at all, so the widely repeated assertion that they run in the microtesla range is unverifiable as stated — though one documented amateur PEMF build reaches about 10 gauss, roughly 1,000 µT, which would be some nine orders of magnitude above the natural signal [documented as a hobby build, not as the commercial class]. The definitional point does not depend on the number. Whatever such a device emits, it is not the Schumann resonance. It is an artificial field wearing its number. Clinical evidence remains preliminary and placebo-confounded [contested], and nothing here is medical advice.

Verdict. Something continuous and planetary is genuinely there — four somethings, in two unrelated physics, none of them audible and none of them steady. What is not there is any channel by which a person could be in tune or out of tune with them. And the intuition underneath — that we are living inside a larger ongoing motion that does not stop when we stop attending to it — is not refuted by a single line above. It does not need a picotesla to stand up. It is simply not the kind of thing a picotesla measures.

13. Is anything broadcasting 440 Hz at the world?

This is the question people usually mean when they say the world has been retuned: not a decision in a committee room, but a signal, everywhere, all the time. It deserves a direct answer, and we owe the reader an account of how we got the answer wrong first.

The answer is no. No documented continuous emission of human infrastructure, at global or near-global scale, is at 440 Hz [documented]. That is the claim we will defend.

The absolute version — that nothing anywhere emits 440 Hz continuously — is unverifiable and should never be made, because it is trivially false. A signal generator does it. A telephone left off the hook does it. Bounded claims survive; unbounded ones invite a rebuttal from a bench in a physics lab. And notice the shape of the gap once the bounded claim is in place: the one near-global continuous channel sits below 440 Hz by about nine-tenths of an order of magnitude, and the one genuinely global channel sits above it by about six and a half [calculated]. Nothing standing sits at it.

Our own correction, stated before the evidence. An earlier version of this front led with the confident line that the 50/60 Hz grid is the continuous global hum, and an adversarial reviewer found it overstated on four counts. The coverage figure was derived from a multi-tier electricity-access statistic that counts off-grid solar — hundreds of millions of people on direct current with no mains frequency at all. The audible channel turned out to require plant within earshot and could not be established as universal. Calling a picotesla-to-microtesla magnetic field a "hum" fused two different things under one word. And the verified channel count was one, not two. The mechanism of that error has a name, and it is the same one this study spends its length warning about: inference by adjacency. Two true statements standing next to each other were read as one claim. The corrected version is below, and it is less dramatic and more useful.

The 50/60 Hz grid: one verified channel, not two

The audible part first, because it is the part that is not global. Transformer cores expand and contract under an alternating magnetic field — magnetostriction — twice per cycle. So the acoustic fundamental is 100 Hz on a 50 Hz grid and 120 Hz on a 60 Hz grid, with harmonics at integer multiples of that fundamental arising from core non-linearity [documented].

Which settles a question worth settling. 440 Hz is not a harmonic of either grid. 440 ÷ 100 = 4.4; 440 ÷ 120 = 3.67 [calculated]. On a 50 Hz grid the neighbouring harmonics are 400 and 500 Hz. On a 60 Hz grid they are 360 and 480 Hz. There is no arithmetic relationship to find, and an earlier draft of ours reached for one by dividing 440 by 50 and 60 rather than by the acoustic fundamentals it had just established — which is the same error twice in one paragraph.

Sound-level limits for this equipment are set by NEMA ST-20 for dry-type units, covering ratings up to 1000 kVA and 1.2 kV, running from roughly 40 dB below 9 kVA up to roughly 64 dB at 701–1000 kVA; liquid-filled limits live in NEMA TR-1, a different document [documented]. IEEE C57.12.90 is a test code — it tells you how to measure, not what the level is, and citing it for a decibel range or for the magnetostriction physics is a misattribution we made and have removed. Any claim that this hum is universally audible is unverifiable: it requires plant within earshot, and most of the world's population is not within earshot of a transformer at any given moment.

The magnetic part, which is the one that survives. WHO Fact Sheet 322 (2007) gives average residential power-frequency magnetic fields of 0.07 µT in Europe and 0.11 µT in North America, and of the order of 20 µT directly beneath high-voltage transmission lines with electric fields of several thousand volts per metre [documented]; other authorities report up to about 30 µT and about 12 kV/m at the extreme, so "of the order of" is the honest phrasing rather than a single figure [documented].

ICNIRP's 2010 low-frequency guidelines set the general-public reference level at 200 µT at both 50 and 60 Hz, raised from the 1998 values of 100 µT at 50 Hz and 83 µT at 60 Hz [documented]. So a typical home sits roughly two to three thousand times below the limit — about 2,900× in Europe and about 1,800× in North America, which is why a single round number was the wrong way to say it [calculated]. Beneath the lines themselves, of the order of ten times below.

IARC classified extremely-low-frequency magnetic fields Group 2B, "possibly carcinogenic", in Monographs Volume 80, published 2002, on a statistical association with childhood leukaemia at higher exposures [documented as a classification]. The two cut-points that circulate should not be merged: the pooled doubling is at ≥0.4 µT (Ahlbom 2000, RR 2.00); at ≥0.3 µT the pooled figure is not significant [documented]. WHO states that the epidemiology is weakened by selection bias and confounding, is unsupported by laboratory and animal work, and lacks a credible proposed mechanism — that WHO says this is [documented]; the association itself is [contested]. Section 11 covers the same ground from the health side and reaches the same place: real, small, unexplained — and, on the 2022 pooled data and the 2024 SCHEER opinion, shrinking.

Coverage, stated in the right unit. Global electricity access reached 92% of population in 2023, with 666 million people without [documented — Tracking SDG7 2025]. Two things follow that our first draft ran together. Access is measured in people, not dwellings. And the grid-connected share is lower than 92%, because 561 million people were served by off-grid solar in 2023 — direct current, with no mains frequency at all [documented].

So the honest summary: this is the single verified continuous near-global channel, and it is magnetic, at microtesla scale, and not a sound. Calling it a hum is a metaphor that costs accuracy, and this study will not spend accuracy on a better sentence.

VLF and ELF transmitters: powerful, few, and not global

ZEVS, the Russian Navy's ELF transmitter on the Kola Peninsula near Murmansk, operates at 82 Hz through a ground dipole of two grounded lines about 60 km long; substations rated 7.478 and 7.446 MVA imply 10–14 MW of input power, and it is described as the most powerful transmitter in Europe [documented]. Stanford's radio-noise systems received it at antipodal distances in the early 1990s, including at Arrival Heights, Antarctica [documented — Fraser-Smith]. Duty cycle and radiated efficiency are not public [unverifiable].

The American counterpart, Project ELF at 76 Hz, ran from Clam Lake, Wisconsin — the Sanguine test facility dates to 1968–69, the first ELF transmission to May 1982, the upgrade to 1985 — paired with Republic, Michigan, completed in 1989, when the system came fully online. Both sites were shut down in 2004 [documented]. An earlier draft dated these to 1977 and 1980; both were wrong and both are corrected.

The United States now runs VLF instead: NAA Cutler, Maine, at 24 kHz, up to 1.8 MW input with 2 MW stated transmission power, continuously encrypted MSK, operational since January 1961. It is not strictly continuous at a single array — an array cannot radiate while it is being de-iced, which is part of why there are two [documented].

None of these is at or near 440 Hz. The closest are 76 and 82 Hz, both roughly five times below it.

Time-signal stations: continuous, but regional — and the one real 440 Hz finding

WWVB (Fort Collins, 60 kHz — NIST defines its coverage by 100 µV/m contour maps at 50 kW, spanning Canada, the United States and Mexico and expanding at night), DCF77 (Mainflingen, 77.5 kHz, official range about 2,000 km), MSF (Anthorn, 60 kHz, with quoted ranges varying from 500 to 1,500 km [contested]), JJY (Japan, 40 and 60 kHz) and BPC (Shangqiu, Henan, 68.5 kHz) all transmit continuously [documented]. None is global.

And here is the one honest 440 Hz emission in this entire section, which we would have missed had we been looking only for confirmation. NIST's WWV and WWVH broadcast a 440 Hz tone — the A above middle C — once per hour: at minute 2 on WWV and minute 1 on WWVH, each standard audio tone lasting about 45 seconds, omitted during the first hour of each UTC day [documented — NIST].

That is a real, deliberate, standards-body 440 Hz emission. It is also about 45 seconds per hour, on shortwave, with propagation that varies by time of day and solar conditions, from two stations. Hourly is not continuous. Shortwave is not global. It is the closest thing in existence to what people imagine, and the distance between it and what people imagine is the whole point.

Broadcast, cellular, Wi-Fi

Continuous carriers, certainly — but spanning roughly 0.5 to 1.7 MHz for AM, 87.5 to 108 MHz for FM, and up to about 6 GHz for cellular and Wi-Fi at 2.4, 5 and 6 GHz — and cell-bounded by design rather than global [documented]. At 2.4 GHz the separation from 440 Hz is a factor of about 5.5 million [calculated]. An earlier draft of ours wrote this band as "0.5–6 GHz," which puts AM radio three orders of magnitude away from where it lives.

The near-miss worth naming, because it is the strongest version of the question

400 Hz is a genuinely standardized power frequency. MIL-STD-704, first issued in October 1959, specifies 115/200 V at 400 Hz for military aircraft; naval vessels run 400 Hz busses; airport ground power units convert 50/60 Hz to it [documented]. It is the closest widely standardized power frequency to 440 Hz, and it exists for a reason that has nothing to do with music — higher frequency allows smaller, lighter transformers and motors, which matters when the thing has to fly.

It is still not 440 Hz. It is not global. And it is not continuous at any fixed point on the ground.

What actually fills the role

Genuinely global and continuous: satellite navigation. Every GPS satellite transmits continuously on L1 = 1575.42 MHz; nominal transmitted C/A power is 26.8 dBW, about 479 watts, and the minimum received power is specified as −158.5 dBW into a 3 dBi antenna above 5° elevation [documented — IS-GPS-200]. An earlier draft called 26.8 dBW a "minimum EIRP"; it is a nominal transmitted figure, and the minimum is a received specification, which is a different quantity entirely. Five to eight satellites are visible from any point on Earth at any moment [documented]. That is what a genuinely global continuous emission looks like, and it is 3.6 million times higher in frequency than 440 Hz.

The reference underneath the timekeeping: 9,192,631,770 Hz, the caesium-133 hyperfine transition that defines the SI second [documented]. VLF keying, WWVB's time code and GPS L1 are all disciplined to it.

But not the grid — and this is a correction we owe, because the tidy version was wrong. Grid frequency is set by load-frequency control, and NERC retired the manual time-error-correction standard, so the North American mains is not caesium-disciplined even in long-term average [documented]. The sentence "everything above derives from caesium" was inference by adjacency: true of the time and navigation signals, false of the one channel we were most interested in.

The nearest 440 Hz gets to ubiquity in human infrastructure remains something small and domestic: the North American dial tone, 350 + 440 Hz at −13 dBm, and ringback at 440 + 480 Hz at −19 dBm, under the Bell Precise Tone Plan, held to ±0.5% in frequency and ±1.5 dB in amplitude [documented]. Mass-scale, precise, and standardised. Also on-demand, in-band audio, and regional.

The honest reframe

So there is no transmitter. There is no tone. Nobody is beaming a pitch at anybody, and the sensation that something continuous surrounds you is not produced by one.

But the sensation is not wrong either, and it would be a poor study that left a person there with nothing. Something continuous does surround most people, and it hums at 50 or 60 — not 440. It reaches the great majority of the world's population, it is magnetic rather than acoustic, it sits two to three thousand times below the exposure limit in an ordinary home, and its one open health question is a small, forty-year-old association with childhood leukaemia that no mechanism explains and that the newest and largest pooled data do not reproduce. That is the true version of the thing being reached for. It is quieter than the story, and it is checkable.

And then there is the part that is loud, that is measured, and that this study has spent twelve sections on. The harm we can actually measure from our sound environment is noise — loudness, at night, for years, in a room somebody did not choose. Not tuning. A person who arrived asking whether something is broadcasting 440 Hz at the world was asking a good question. The answer is no, and the reason the question felt urgent is sitting in section 4, measured in decibels, with the receipts.

14. Who decided on 440?

Which leaves the last honest question. If nothing is broadcasting it, where did 440 come from, and who chose it?

The short answer is that the paperwork is public, the dates are exact, and the story it tells is duller and stranger than the one in circulation. What follows is a companion to the 432 Hz study (slug 432-hz), which handles the acoustic and physiological claims; this section handles only the documents.

The question, answered cold

Two instruments in this chain were legally binding, and both of them set A = 435. France's ministerial act of 1859, and the Vienna Convention of 1885 — which was kept in force for Germany by Article 282 of the Treaty of Versailles. Everything that actually produced A = 440 was voluntary: American trade practice, an American standards recommendation, a conference in London, and eventually ISO.

An earlier draft of this section led with the line that "exactly one instrument was ever legally binding," and then contradicted itself two paragraphs later by describing a convention revived by treaty. A convention binds its parties. Two, not one — and the superlative was doing rhetorical work the evidence did not authorise.

1859 — France, and the only national law in the story

A commission on a uniform French pitch was created by ministerial order of 17 July 1858, with Halévy as rapporteur and the physicist Lissajous as its technical authority [documented]. The instrument of 16 February 1859, issued by the Minister of State, Achille Fould, fixed the diapason normal at 870 simple = 435 double vibrations per second at 15 °C and placed a type-fork in state keeping [documented]. It bound France's subsidised theatres, conservatoires and musical establishments [documented].

Forks made or sold in France were verified against the type-fork using Lissajous's optical comparator [documented]. Whether that verification extended to instruments is unverified — our draft said "forks and instruments," and only the forks are documented. Whether the act is properly a décret or a ministerial arrêté, and whether the standard fork sat at the Conservatoire de musique or the Conservatoire des arts et métiers, are reported inconsistently across sources [contested]. The frequently repeated claim of an 1884 Italian decree at A = 432 is dealt with in the next section; it was not verified in this pass to the level of a named ministry.

1885 — Vienna, and a treaty

An international pitch conference met at Vienna from 16 to 19 November 1885, at the instigation of the Austrian Minister of Culture and Instruction, and resolved unanimously on a′ = 435 as international pitch [documented]. Italy, Austria, Hungary, Prussia, Russia, Saxony, Sweden and Württemberg were represented at the conference [documented] — which is a weaker and more accurate statement than "came to use it," and the distinction matters, because how far each state then enforced it is unverified.

Britain and the United States stood outside. Britain instead recognised A = 439 in 1896, derived from a misreading of the French standard as an oboe pitch at 59 °F [documented].

Then the Vienna instrument acquired treaty status, which is the detail almost nobody knows. Article 282 of the Treaty of Versailles, item (22), lists the "Convention of November 16 and 19, 1885, regarding the establishment of a concert pitch" among the conventions that "shall alone be applied as between Germany and those of the Allied and Associated Powers party thereto" [documented — verified at Wikisource]. From 1920, therefore, Germany's only treaty-level pitch obligation ran to 435.

We flag that as suggestive rather than probative. That it cuts against the story told later in this section is an inference we are drawing, not a documented finding, and we would rather say so than let a good line do work it has not earned.

1926 to 1939 — the drift to 440 was American first, then technical

The American music industry reached an informal 440 in 1926, and the American Standards Association recommended A = 440 in 1936 [documented in secondary sources — no primary ASA document was located, and that gap is real].

The documented technical argument for 440 over Britain's 439 is beautifully mundane: 439 is a prime number, and could not be generated by the crystal-divider chains that broadcasting had begun to rely on for a stable tuning reference [documented]. That is the reason. Not a philosophy of consciousness. A frequency divider.

The claim that Sir James Swinburne carried that argument for the Musical Association at a preliminary meeting in 1938 is [contested] — the source usually cited for it does not contain it, and near-identical wording appears in a 2017 music-journalism feature. We had it tagged [documented]. It is not.

And there was no 1938 international conference. The first edition of New Grove (Lindley et al., "Pitch," vol. 14, Macmillan, 1980, p. 785 — noting that our own source's prose says 1981 while its bracketed citation says 1980) placed a British Standards Institution endorsement of 440 at a conference in May 1938. The right month, the wrong year, and a national body rather than an international one. Studies that cite "the 1938 conference" are copying that error forward, and we nearly did too.

11–12 May 1939 is the real date. The conference met at Broadcasting House, hosted by the BBC, organised by the British Standards Institution under the auspices of the ISA — formally the International Federation of the National Standardizing Associations, though "International Standards Association" was in contemporary use. France, Germany, Great Britain, Holland and Italy sent delegates; Switzerland and the United States submitted views in writing; the international broadcasting and telephony bodies were represented. Sir Cecil Graves, deputy director-general of the BBC, formally opened it; G. W. C. Kaye presided. The resolution was that the standard "shall be based on a frequency of 440 cycles per second for the note A in the treble clef" [documented — G. W. C. Kaye, "International Standard of Concert Pitch," Nature 143, 905–906, 27 May 1939, read through a full secondary reproduction rather than the paywalled original].

One thing our draft had exactly backwards, and it is worth naming because it was the load-bearing sentence. We wrote that the BBC was already radiating a 440 tuning tone before any international meeting. Our own source says the opposite. Broadcasting pushed for standardisation through the 1930s, and the BBC began broadcasting the A = 440 tuning note after the May 1939 conference [documented]. Reversed, and corrected.

War came in September. The ISA ceased activity in 1942 [documented]. Whether an ISO endorsement intervened in 1953 is reported in secondary sources and unverified; the silent jump from 1939 to 1955 in most accounts, including our first draft, is a hole rather than a continuity.

1955 and 1975 — ISO, and what "standard" means legally

ISO took the recommendation up in 1955 as ISO Recommendation R 16 [documented]. ISO 16:1975, Acoustics — Standard tuning frequency (Standard musical pitch), first edition 1975-01-15, ISO/TC 43: the frequency for A in the treble stave "shall be 440 Hz," and "tuning and retuning shall be effected by instruments producing it within an accuracy of 0,5 Hz." It replaces R 16-1955, "to which it is technically identical" [documented]. Note the comma — that is the ISO abstract's own decimal notation, and the "±" that usually gets attached to it is an interpolation, including ours.

On the standard's current status our own fronts disagree, and we will show the disagreement rather than pick. One front read the catalogue entry as "reviewed and confirmed 2022, current." Another, reading the same record through a mirror because iso.org returned HTTP 403, found a confirmation cycle opened 30 September 2022 with a stated completion date of 14 February 2026 — a date now past, which makes the live status unverified. Anyone citing ISO 16 should open the catalogue page themselves.

ISO is a non-governmental organisation and its standards are voluntary, binding only where a legislature incorporates them or a contract requires them [documented]. Exactly one national law mandating A = 440 surfaced in this study — Italy's, in 1989, and the next section sets it out. Whether it is the only one we are keeping unverified rather than promoting to a finding, because that would be an absence of evidence over a search we know to be incomplete: mandatory state standards in planned economies were not checked. Orchestras playing at 442 or 443 are breaking nothing anywhere, including in Italy, where the duty falls on institutions and on the calibration of tuning devices rather than on performance.

The Goebbels claim, and where it comes from

Now the part people arrive for. The story that A = 440 was imposed by Nazi propaganda has a traceable origin, and tracing it is more interesting than either believing it or sneering at it.

The origin is Laurent Rosenfeld, "How the Nazis Ruined Musical Tuning," Executive Intelligence Review 15:35, 2 September 1988, p. 54 [documented] — a publication of the LaRouche movement, whose Schiller Institute launched a campaign for A = 432 as "Verdi tuning" in 1988, with a symposium at the Casa Verdi in Milan on 9 April 1988 [documented]. Rosenberg's peer-reviewed study (Journal of Popular Music Studies 33:1, 2021, 137–154) is cited as identifying that article as the root of the internet claim [documented as to the citation — we could not read Rosenberg; the publisher returned HTTP 403].

Secondary accounts report Rosenfeld claiming that Radio Berlin, Goebbels's mouthpiece, organised the 1939 conference, and that the French were not invited. We could not read the original — the EIR archive fails TLS verification — so the widely repeated description of Rosenfeld as "director of the Acoustic Committee of Radio Berlin" is unverified, and we have removed our paraphrase of it rather than pass it on.

But the checkable part is checkable, and the record refutes it. France sent delegates. The conference was in London, convened by a British standards body, opened by a BBC executive, and presided over by a British physicist [documented]. And the timing is decisive on its own: the American Standards Association recommendation predates the conference by three years, and American industry practice predates the regime entirely [documented].

Which brings the plainest fact of the section, and it should be said in one line. Pitch had been standardising for eighty years before 1939 — from the French commission of 1858 and the arrêté of 1859, through Vienna in 1885, through British 439 in 1896, through American 440 in 1926, to the ASA in 1936. The 1939 meeting was the late stage of a slow, public, thoroughly documented process, not its beginning.

The mechanism, and it is not a people. What produced this story was a political campaign that needed an origin villain for a standard it opposed. That is a mechanism — a rhetorical structure that recurs in many campaigns, on many subjects, in many countries — and it is not a nation and not a people, and this study will not let it become one. Rule 4 applies here as hard as anywhere in this document.

And a word for the person who repeated it. Almost everyone who has passed this claim along is four or five hands removed from a 1988 pamphlet they have never seen and could not be expected to find. There is no shame in that whatsoever. It is what a well-told story does; it is what all of us do with most of what we know. Being handed the provenance is not being caught out. It is being handed something better than the story was.

15. What the world actually tunes to

Here is the practical follow-up, and the answer changes the shape of the argument more than any document does.

A pitch standard is real, and in one country it is law. What the law binds is the tuning fork, not the orchestra.

Has any country ever adopted 432?

Yes — Italy, by decree in 1884, superseded internationally within about a year. And the same country has mandated 440 Hz by statute since 1989. That second sentence is the one that never travels with the first.

The Italian Wikipedia entry states that "il governo italiano emise un decreto per la normalizzazione del diapason a 432 vibrazioni per secondo, nel 1884," following an 1881 proposal. Our draft attributed this to a "War Ministry decree." No source we reached names any ministry, and the attribution is demoted to unverifiable. Nor did we locate any repeal instrument, so "lasted about a year" becomes the more careful "superseded internationally within about a year" — Vienna 1885 settled on 435, and Italy was represented there.

On Verdi himself, two cited sources say different things, and we will print both rather than choose. Il Sole 24 Ore reports that he officially requested adoption of "the A with 432 vibrations"; the Italian Wikipedia entry has him backing 435 for the Requiem [contested]. What survives both accounts, and is the historically interesting part: the 432 figure came from the government commission, chosen for arithmetic convenience — 432 divides neatly in a way 435 does not — and Verdi described the difference between the two as imperceptible. No source from 1884 contains any cosmological argument whatsoever.

That last point is worth sitting with, because the campaign that revived 432 proves it by accident. The Milan symposium of 9 April 1988 included a paper on "the astrophysical basis of the C = 256 tuning." Verdi's 432 was arithmetic. The cosmos was supplied a century later, by people who needed the number to mean more than it did.

And Vienna 1885 itself is a small lesson in checking sources. One reference states that the conference established a′ = 435 as the standard across Italy, Austria, Hungary, Russia, Prussia, Saxony, Sweden and Württemberg. The Schiller Institute's own history states that the conference "ruled that no such pitch could be standardized." Two cited sources, opposite claims, on a conference 140 years ago. We report the conflict [contested] and note which reading each side needs.

Italy's actual law, which is at 440

Legge 3 maggio 1989, n. 170, "Normalizzazione dell'intonazione di base degli strumenti musicali," published in Gazzetta Ufficiale Serie Generale n. 109 on 12 May 1989 and in force from 27 May 1989 [documented]. Article 1, verified verbatim: "Il suono di riferimento per l'intonazione di base degli strumenti musicali è la nota La3, la cui altezza deve corrispondere alla frequenza di 440 hertz (Hz), misurata alla temperatura ambiente di 20 gradi centigradi." Article 2 saves research and artistic requirements. The duty falls on state-subsidised musical institutions and orchestras and on the public broadcasting concessionaire. The sanctions attach to reference tuning devices — forks, plates, electronic generators — not to performances [Articles 1 and 2 and the broadcasting duty: documented, verified directly. The remaining articles and the penalty amounts: contested — read in secondary reproduction only].

And the bill and the statute are the same instrument. The 432 bill is Senate no. 1218 of the Tenth Legislature, communicated to the Presidency on 20 July 1988, signed by Boggio, Mezzapesa, Cappelli and Azzarà among others. Its Article 1: "…la nota La3, la cui altezza deve corrispondere alla frequenza di 432 Hertz (Hz), misurata alla temperatura ambiente di 20 gradi centigradi." Same title, same sentence, same saving clause, same structure — one number changed [documented as to Article 1 on both sides]. The bill did not fail. It passed, at 440, ten months later.

Two things must not be read into that. The bill's own explanatory report says its text was taken from an earlier bill, no. 296, which proposed 440 — so the borrowing may run the other way, and the causal question is unresolvable from the public record. And nothing here is anybody's conspiracy: it is a parliament reusing a drafted text and settling on the number the rest of the world already used. What it does establish is the flat fact our draft lacked: a bill for 432 in July 1988, a statute for 440 within a year, in the same words.

It also forces a correction to our own thesis sentence. We wrote that "the standard exists; the compliance does not," and that "no orchestra is audited." That is overstated. ISO 16:1975 fixes device accuracy at 0.5 Hz, and Italy enforces exactly that in statute. The accurate formulation is narrower and more interesting: the standard binds the tuning fork, never the ensemble.

What ensembles actually play at

Nothing like a single number, and nobody stops them.

Two entries in that table are weaker than we first wrote them, and both corrections point the same way. Figures in circulation for Boston at 441 or 444, and for Vienna at 443, 444 or 447, are [contested] or [unverifiable]. And the New York Philharmonic at 442 is sourced only to Bernstein-era practice, not to any current policy — writing it in the present tense was our error. Berlin at 443, down from 445, is the one that holds up verbatim [documented].

Note what the spread means. If a semitone is 100 cents, the distance from 432 to 445 is about 51 cents — half a semitone — and it is spanned by working professional orchestras right now, with no legal consequence and no reported epidemiology. The 432-to-440 gap is 31.77 cents, and it sits inside a range the profession already uses.

The 1988 petition, honestly counted

The Schiller Institute's petition for a lower tuning is real and its archive is public. Our draft made three claims about it that do not survive.

We wrote that the campaign's own archive concedes the proposal did not become law. It does not. The archive page says nothing about the outcome, and nothing is conceded. The clause is deleted and replaced by the documented fact that matters more: Italy legislated 440 the following year.

We gave a signatory total of "over 2,000 musicians." The archive gives no total — it offers a full list on request. The number is cut. Named messages of support attributed to particular singers are likewise unverified and are not repeated here; the petition had genuine and distinguished support, and that can be said without inventing a count.

And a musician sometimes cited as having released a 432 album — the source we reached names an album, unnamed, with no year. That is all we can say.

The premise, quietly dissolved

Now the thing that changes the argument, and it is not a rebuttal so much as a wider window.

Indian classical music does not fix a reference pitch at all. In both the Hindustani and Carnatic traditions the tonic — Sa — is not a frequency. It is a position, chosen to suit the voice of the singer or the build of the instrument, and everything else in the raga is defined in relation to it. The tanpura or the electronic śruti box is tuned to whatever Sa has been chosen for that performance, by that performer, on that day. Two singers may perform the same raga a whole tone apart and neither is transposing, because there was no absolute reference to transpose from [documented as musical practice; we did not trace a primary musicological citation in this pass, and any specific frequency values circulating for particular voice types are unverified here].

Sit with what that implies. A tradition of enormous sophistication, with a theory of intervals and consonance far older than the Vienna conference, and a devotional repertoire of extraordinary depth, simply does not have the parameter the argument is about. There is no wrong tuning to be rescued from and no right one to return to. The premise — that humanity was moved off a correct reference — has no purchase in a tradition that never fixed one.

The same holds, in different ways, wherever the tonic is chosen rather than given. And it is worth noticing that the traditions most often invoked as evidence that the ancients knew something we have forgotten are frequently the traditions with no fixed reference pitch at all. What they had was not a better number. It was a different idea of what a note is for.

16. Which devices are set to 440, and which have no pitch at all

The last practical question, and the one that quietly settles the matter for anyone trying to act on the belief: if you wanted to live at 432, what would you actually have to change?

The arithmetic first, because it is small and exact

432 ÷ 440 = 0.981818. That is −31.77 cents, or 1.818% [calculated — 1200 × log₂(432/440)]. Resampled rather than pitch-shifted, a three-minute track becomes 3 minutes 3.3 seconds [calculated].

Thirty-two cents is about a third of a semitone. It is clearly audible as a difference when two versions are compared back to back, and it is inside the range that working orchestras already span between 432 and 445.

What carries a tuning reference, and what does not

Two device details worth having if you own the hardware: the Korg CA-50 loses its calibration memory when the batteries are replaced, and the Yamaha P-71 has no Master Tune menu at all — it is a hidden key combination, [FUNCTION] held with G♯6 or A6 [documented]. Model lists for other units in our draft (Korg CA-1 and CA-40; Yamaha CVP-701 and reface) are unverified and have been dropped rather than kept as plausible.

One correction we owe the reader in full, because it is the clearest example in this study of the error we keep naming. We cited Apple's Logic Pro documentation as "stating the limit plainly" — that global tuning does not affect audio regions or external MIDI. Apple's page says no such thing. It says only that the parameter "determines the global tuning of all software instruments," that the default is concert pitch A (440 Hz), and that detuning is in cent steps. The exclusion is a correct inference from the documented scope of the parameter. It is not a quoted denial, and presenting it as one was inference by adjacency committed inside a section about inference by adjacency. Corrected: quote [documented], exclusion [inference].

Similarly, a popular retuning app was cited by us for the proposition that it cannot reach streaming services. Its own site says the opposite — it advertises retuning of major streaming platforms and thousands of radio stations inside its own app, and sells a browser extension. That is a vendor claim, unmeasured [contested], and we are not in a position to confirm or refute it. What survives from our original point is only the narrower and still-true part: there is no operating-system-level path on iOS or Android. Each app must do it itself.

What real-time retuning would actually cost

There are two DSP families, and each costs something different. There is no third — with one caveat noted at the end.

Resampling. Play the samples out at a different rate. Cheap, clean, and it changes the speed along with the pitch: the 1.818% that turns three minutes into 3:03.3. For anything with a beat, a groove or a synchronised video, that is not a neutral change.

Time-domain or phase-vocoder pitch shifting. Hold the duration, move the pitch. This costs two things. Latency, because the algorithm works on windows: the Rubber Band library's integration documentation gives window sizes of 512, 1024 and 2048 samples for its R2 engine and 1280 and 2048 for R3, which is roughly 11 to 46 ms at 44.1 and 48 kHz [documented plus arithmetic — an earlier draft gave "~30–50 ms" and cited the wrong document]. And artefacts: transient smearing and phasiness, worst on percussion and on solo voice. The better-sounding engine is also the slower one — the library's developer reports R2 running about three times as fast as R3 [documented], which corrects our draft's "3–4×."

Two practical notes. On the desktop, a genuine system-wide pitch shift is not a solved problem: the Equalizer APO project's request for pitch support is ticket #9, opened 4 August 2015, last updated 12 December 2015, with no developer reply [documented]. That evidences the absence of the function, which is all we should claim; the workaround people describe, chaining a VST host through a virtual audio cable, is [contested — forum practice] and not something we verified. And in a DAW, alternative-tuning support is narrower than it looks: Ableton Live 12 loads .scl and .ascl tuning files, and they reach Live's own instruments and only those MPE-compatible VST2, VST3, AU and Max for Live devices with per-note pitch bend set to ±48 — never audio clips [documented]. Our draft called this "contested across forum threads"; it is not contested, it is documented and simply narrow.

Two commands anyone can run to hear the difference, both reproducible rather than sourced [calculated]:

The media-player recipe that circulates — setting playback rate to 0.981818 with time-stretching enabled — rests on a forum thread we could not read, so the specific recipe is [contested — user guides]. The concept is sound: the "enable time stretching" preference exists precisely to hold pitch when rate changes, so disabling it makes rate change pitch. Measure it rather than trust it.

And the caveat on "no third family": stem separation followed by per-stem retuning does exist. It is not a third approach so much as the first two applied more expensively — it inherits both the latency and the artefacts, and adds separation artefacts of its own.

The 31.77 cents in perspective

One number ends this section. Consumer digital audio clocks are typically specified to about ±50 parts per million, which is 0.0866 cents [calculated]. The 432 retuning is therefore about 367 times larger than the worst clock error your equipment is allowed to have [calculated]. Whatever Bluetooth is doing to hold two clocks together — and the mechanism varies between asynchronous sample-rate conversion and sample insertion or dropping, so we will not generalise [contested] — no consumer clock error approaches 31.77 cents.

Which is the honest bottom line of the whole practical question. The change is real, it is measurable, it is easily audible in a direct comparison — the laboratory pitch-discrimination thresholds are a few cents for musicians and under twenty for untrained listeners, so a third of a semitone is not subtle — and it can be made, on your instruments by a menu and on your recordings only by re-rendering the audio, at a cost in either speed or artefacts. Nothing in the acoustic environment is doing it to you either way. And one thing no menu reaches: your room, your speakers and your ear canals keep their own resonances wherever the music is tuned (section 12), so the result is yours and does not travel. It is a preference, and it is allowed to be one.

17. Where 432 Hz actually fits

The companion study, slug 432-hz, does this subject properly across its full length, and this section will not repeat it. What belongs here is only the join: how the tuning claim relates to the evidence assembled above.

The 432 narrative is not one claim but a stack of at least six — an aesthetic claim, a historical claim, an attribution claim, a physical claim, a physiological claim, and a therapeutic claim — presented as though accepting the first commits you to the last. It does not. The companion study finds that some survive scrutiny and others do not, and it names which is which.

What this study can add is the shape of the mismatch.

Every threshold in this document is a sound pressure level in decibels. Road 53 dB Lden. Night 45 dB Lnight. Aircraft 40. Occupational 85 dB(A). The NANR45 curve, 92 dB down to 34 dB. The infrasound threshold, 79 dB SPL at 20 Hz. Not one is expressed in hertz [documented — provable by opening the recommendation tables and reading the units].

That is not a rhetorical flourish, and it is deliberately phrased as a positive, checkable statement rather than a negative one. An earlier draft of this study wrote that "nothing in these documents references frequency," which is a negative claim over documents we had not read end to end — exactly the kind of claim this house is not allowed to make. The restated version is verifiable in a minute by anyone with a browser.

Frequency is not absent from acoustics, of course. It is central: A-weighting is a frequency weighting, the low-frequency section above is entirely about frequency-dependent perception, and G-weighting exists precisely because infrasound needs its own curve. What is absent is any health threshold expressed as a musical tuning reference. The two senses of "frequency" — the spectral content of a noise, and the pitch standard a violin is tuned to — are not the same subject, and the narrative depends on treating them as though they were.

So the honest placement is this. The claim that A=440 concert pitch causes health harm has no place in this evidence base [myth]. And the instinct that brought a person to that claim — that the modern acoustic environment is doing something to them — is correct and documented. The variables are loudness, night-time exposure, chronicity, and lack of control. Someone who swaps a tuning fork for a pair of earplugs and a conversation with their local authority about the ventilation unit next door has not been disillusioned. They have been given something that works.

18. What to do — graded by the evidence, and prescribed to nobody

Everything in this section is a description of what the evidence supports at population level. None of it is advice to any individual, and nothing here should displace a conversation with a clinician about your own body. Some of what follows will be irrelevant to your circumstances and some of it will be impossible in your housing. That is not a failure of the reader.

The measures are graded by how well evidenced they are, which is not the same as how effective they are, and is very much not the same as how popular they are. Notice as you read that the ordering is close to inverse to what the wellness market sells.

Tier 1 — best evidenced

Removing exposure at the source. Every guideline document cited in this study exists to drive source control: quieter road surfaces, traffic volume and speed management, night flight restrictions, rail maintenance, setback distances, and land-use decisions that do not place bedrooms against arterial roads [documented — this is what the WHO and EEA guidance architecture is built for]. This is the highest-leverage intervention and the one an individual has least direct control over, which is an uncomfortable pairing and an honest one.

Hearing protection at genuinely loud events. The one randomised trial: earplugs at a music festival cut temporary threshold shift from 42% of ears to 8%, and new tinnitus from 40% of people to 12% [documented — Ramakers 2016, n = 51, single event, single-blind, and TTS is a surrogate for cumulative damage rather than a measure of permanent loss].

Earplugs for sleep, within their range. In the 2026 SLEEP polysomnography trial, earplugs recovered about 72% of the slow-wave sleep lost to traffic noise, and began failing at 65 dBA [documented]. Effective across the ordinary range; not a solution to a genuinely loud environment. Whether long-term nightly use is net-beneficial is unverifiable — nobody has run that study.

Tier 2 — evidenced by guidance and physics, with the specific quantities unverified here

Building fabric: glazing, façade treatment, and which room the bed is in. This is what the entire regulatory apparatus is designed to influence, and the direction is not in doubt. The specific numbers commonly quoted — barrier attenuation of 6–7 dB in the 500 Hz to 1 kHz band, a "quiet façade" defined at 10 dB(A) or more of difference, a measurable sleep benefit from orienting a bedroom away from the road — are unverified in this pass [unverifiable] and are named as directions rather than quantities we can vouch for.

Escalating a low-frequency complaint properly. If a hum is being dismissed on the basis of a compliant dB(A) reading, the technically correct request is for a third-octave band analysis compared against a low-frequency criterion curve — the UK's NANR45 procedure being one published example, running from 92 dB at 10 Hz down to 34 dB at 160 Hz, with a 5 dB relaxation for daytime-only noise and a further 5 dB for steady sounds [documented]. NANR45 is an assessment procedure, not a statutory limit, and no UK exposure standard for low-frequency noise exists [documented]. Knowing the name of the right measurement is often the difference between being taken seriously and being told the reading is fine.

Time and place rather than device. Nothing in the sleep literature is about equipment. It is about when sound arrives — the +10 dB night penalty in Lden is the empirical finding written into the unit — and about intermittency. A predictable, steady, controllable sound is a different exposure from an unpredictable, intermittent, uncontrollable one, even at the same measured level [documented as the basis for the Lden weighting and the NANR45 steady-sound relaxation].

Tier 3 — popular, and the evidence is weaker than the marketing

White and pink noise masking. GRADE-rated very low quality across 38 articles, too heterogeneous to pool, with continuous noise found both to improve and to disrupt sleep [documented — Riedy 2021]. Self-reported sleep quality improves in pooled trials [documented — Ding 2025], on an instrument the habituation literature shows is unreliable for sound. The one clean objective finding runs the other way: pink noise at 50 dBA cut REM sleep by 18.6 minutes [documented]. And three of fourteen infant sleep machines exceeded 85 dBA at 30 cm at maximum volume, with all fourteen exceeding the 50 dBA nursery limit [documented — Hugh 2014].

This is not a prohibition. It is a statement that masking is the weakest link in the chain rather than the strongest, and that at high volumes close to a small child it is itself an exposure.

Binaural beats. Genuine perceptual phenomenon [documented]. Cortical entrainment contradicted more often than supported [documented]. Small-to-moderate effects on anxiety and attention across independent meta-analyses, g around 0.40 to 0.45 [documented]. Perioperative anxiety effects large but with I² = 91.6%, which means the pooled magnitude should not be quoted [contested]. Compatible with expectancy and attentional distraction, because no retrieved trial was blinded to expectancy [documented limitation].

Active noise cancellation. Commonly described as effective mainly against steady low-frequency sound and poorly against speech, and sometimes credited with lowering the volume people choose for their music. Both specifics are unverified here [unverifiable]. The mechanism is real; we could not source the performance claims.

Music as an adjunct in clinical settings. Genuinely evidenced, and worth taking seriously — perioperative pain SMD −0.77, anxiety −0.68, analgesia use −0.37 [documented]; rhythmic auditory stimulation for stroke gait, velocity effect sizes 0.68 to 0.98 across two reviews [documented]; music therapy added to usual care for depression, clinician-rated −0.98 [documented]. Every one of those is adjunctive, added on top of standard treatment, never instead of it.

Tier 4 — not supported

Singing bowl therapy: four peer-reviewed studies exist, and the reviewers state they cannot recommend it at this stage [documented — Stanhope & Weinstein 2020]. That means the studies to decide have not been done, not that it has been disproved.

Organ-specific "healing frequencies," tuning-based health claims, and devices sold to restore a Schumann connection: [myth], and in the last case not supported by any measurement study we could find [unverified].

Infrasound as the explanation for residential symptom clusters: the null is better supported, and the documented low-frequency problem is audible [documented]. Whether sub-audible infrasound at real-world levels causes any health outcome is unverifiable — few studies are powered to detect small effects, and that is a genuine gap rather than a closed question.

What would change these answers

Four studies would move this field more than another decade of the current kind:

An intervention trial. No randomised trial has ever tested whether reducing noise reduces cardiovascular events [documented]. Until one exists, every causal claim in section 5 is inference from observation.

A clean separation of noise from traffic-related air pollution. The two co-vary almost perfectly and both co-vary with socioeconomic position [contested]. Every burden figure in this study inherits that uncertainty.

Personal, indoor, longitudinal exposure measurement. Every exposure figure above is modelled at the façade, ignoring window orientation, glazing, which room people sleep in, and whether they open the window [documented]. That misclassification biases toward the null and may be hiding a real effect.

An honest resolution of the disability weight question. Annoyance is a self-reported state carrying a weight outside the standard Global Burden of Disease set, and it plus sleep disturbance are about 92% of the headline total [calculated]. Germany's national study shows one weight choice can double a burden estimate [documented]. Until this is settled, "one million healthy life-years" is a defensible policy figure and an indefensible scientific constant.

19. The mechanism, and the closing thought

It would be easy, and completely wrong, to end this study by naming someone.

There is no villain in the evidence above. There is no industry that decided people should not sleep, no authority that chose harm, no group whose fault this is. What there is, instead, is a mechanism — and mechanisms are harder to be angry at, which is precisely why they survive so long.

The mechanism has four parts, and none of them required anyone to intend harm.

Sound was classified as a nuisance rather than an exposure. Nuisances are managed by complaint. Exposures are managed by limits, monitoring and enforcement. Air quality made that transition decades ago; sound largely has not. This single categorical choice explains most of what follows.

The instrument was built for the wrong organ. A-weighting was designed to approximate the ear's sensitivity, and it does that reasonably well. But the documented harm is cardiometabolic and cognitive, not auditory — and dB(A) attenuates 70 decibels at 10 Hz [documented]. A regulatory system that measures with a hearing-shaped instrument will systematically under-detect a stress-shaped harm, and will do so with complete procedural correctness.

The exposed population is the population that stopped reporting. Autonomic response habituates far less than conscious annoyance does [documented]. So the survey instrument gets quieter exactly as the exposure gets longer, and the people with the most exposure become the least visible in the data. That is not deception by anyone. It is a measurement system running against the biology it is trying to measure.

And the decisions that set exposure were made where nobody was thinking about sleep. Road layout, floorplate economics, glazing specification, plant location, flight paths, the acoustic treatment of an office understood as an aesthetic line item rather than a health parameter. Each of those decisions was rational in its own frame. None of them had a person's slow-wave sleep in the frame at all.

That is the whole of it. No enemy. Four ordinary failures of category, instrumentation, feedback and scope, compounding across a century of building.

Which leaves the closing thought, and it is a simple one.

Almost every intervention this civilisation is enthusiastic about involves adding something. Add a tone, a supplement, a device, a subscription, a frequency. The evidence for the additive interventions in this study, taken honestly, ranges from modest to absent — and the evidence for the subtractive one is the largest body of work in the entire field.

Quiet is not the absence of a treatment. It is the treatment. A room where the body is not being asked, all night, to prepare for something. Twenty-three minutes of slow-wave sleep, given back. A blood pressure that is not being nudged upward six millimetres at a time by something the sleeper will never remember hearing.

It has no manufacturer, no supplement aisle, no subscription, and no one to advertise it. It is available to a great many people who have never once been told to try it, and it is unavailable — through housing, through road placement, through poverty — to a great many others, which is its own quiet injustice and the reason source control sits at the top of the list rather than earplugs.

The feeling that brought you here was right. The world is doing something to you through sound. It simply is not doing it through the tuning of a violin.

It is doing it through loudness, at night, for years, in a room you did not choose.

And quiet — real, boring, unmonetised quiet — is a medicine that costs nothing, and that almost nobody takes.

Corrections log

These are our errors, caught by our own adversarial reviewers before publication. They are listed because a study that asks for care has to demonstrate it.

Second round — from a cold verification pass, 29 July 2026

A check run against this study by a process given no conclusions to defend. Its findings on this article were fewer than on our unpublished working briefs, and two of the three are omissions rather than misstatements — which is the harder kind to catch, because nothing on the page looks wrong.

What this study could not verify

Listed so that nobody has to guess which gaps we know about.

Requiring a document opened by hand: the WHO 2011 per-outcome DALY breakdown beyond sleep disturbance and annoyance — that is, the IHD 61,000, cognitive impairment 45,000 and tinnitus 22,000 figures, and WHO's reason for the "at least" wording; the WHO 2018 leisure-noise value of 70 dB L_Aeq,24h, widely cited but confirmed by no source we reached. The WHO 2011 record at iris.who.int returned HTTP 403 to automated fetch, which is not proof of a dead link but does mean the source list should not be presented as verified-reachable.

Design details and sample sizes: Herzog et al. 2019 sample size (reported as 70, sourced only to a review that misstates a neighbouring study's n); Basner 2011's exposure design; the NHMRC screening counts; the Health Protection Agency 2010 wording on vibroacoustic disease; Møller & Pedersen on perceptibility of individual cycles below 10 Hz (paywalled); the 1993 Taos Hum investigation; Health Canada's hair cortisol and blood pressure nulls.

Effect estimates: the emergency-department music review's trial count and effect sizes; Hole 2015's heterogeneity and publication-bias statistics; Szalma & Hancock heterogeneity; the Granada cardiovascular admissions study (no DOI supplied); Stanhope & Weinstein's PEDro quality range; Oster 1973 in full text.

Mechanistic claims resting on review articles rather than primary papers: eNOS-knockout injury severity; circadian gene dysregulation (Bmal1, Cry1, Per2); the ~10 pT Schumann transient figure; the explanation that public spectrograms encode amplitude as colour.

Structural: whether global ambient sound has risen over the last century — unverifiable, because no long-run comparable measurement series exists. The EEA's 3%-since-2017 figure is a European trend over five years, not a history. Anyone who tells you the world is measurably louder than it was in 1900 is extrapolating.

Added by the five extension fronts. Rosenberg's Journal of Popular Music Studies article (the peer-reviewed provenance study) — the publisher returned HTTP 403. Rosenfeld's 1988 Executive Intelligence Review article itself — the archive fails TLS verification, so the description of its author as "director of the Acoustic Committee of Radio Berlin" is unverified and we do not repeat it. The primary American Standards Association document of 1936. A reported ISO endorsement in 1953, which leaves a silent gap between 1939 and 1955. State-by-state enforcement after Vienna 1885. The text of Italy's Legge 170/1989 — Article 1 is now verified verbatim against three independent Italian reproductions, and Article 2 and the broadcasting duty are confirmed; the remaining articles and the lire amounts in the penalty provision are still secondary only, as is the article-by-article mapping onto Senate bill 1218 of 1988 (whose Article 1 is verified). Neither the Gazzetta Ufficiale text nor the Senate PDF responded to automated retrieval. The identity of the ministry behind Italy's 1884 decree, and any repeal instrument for it. A primary musicological citation for the movable tonic in Indian classical practice, which we state as practice rather than as a sourced claim. The exact ICNIRP low-frequency reference level near 8 Hz (the table would not render). Duty cycle and radiated efficiency for ZEVS. Whether mandatory state standards in planned economies ever mandated A = 440 — not checked, which is why whether Italy's 1989 statute is the only one of its kind stays unverified rather than becoming a result. (That the statute exists is no longer in doubt; only its uniqueness is.)

And a second unresolved disagreement between our own fronts: the live catalogue status of ISO 16:1975. One front read it as reviewed and confirmed in 2022 and current; another, reaching the record through a mirror after iso.org returned HTTP 403, found a confirmation cycle with a completion date of 14 February 2026 — now past. Open the catalogue page before citing it.

And the one we could not settle between our own fronts: whether the EEA's current health estimates are the 66,000 set or the 73,000 set, and whether a corrigendum was incorporated in March 2026. Check the live publication page before citing either.

Sources

Burden and guidelines. WHO Regional Office for Europe / JRC, Burden of Disease from Environmental Noise (2011) — iris.who.int/handle/10665/326424, and the accompanying WHO news release of 30 March 2011. WHO, Environmental Noise Guidelines for the European Region (2018), with the guideline development and GRADE methodology paper (PMC5923855). WHO, Night Noise Guidelines for Europe (2009). WHO, Community Noise Guidelines (1999). Directive 2002/49/EC (the Environmental Noise Directive). EEA, Environmental noise in Europe 2025, its newsroom release of 24 June 2025, thematic briefing 3.3, the exposure-of-population indicator page, and Health risks caused by environmental noise in Europe (2020).

Cardiovascular. van Kempen, Casas, Pershagen & Foraster (2018), PMC5858448. Pershagen et al., Environmental Epidemiology (2025), PMC12122180. Minkin et al., Environment International (2025), doi:10.1016/j.envint.2025.109667. Münzel, Peris & Sørensen, European Heart Journal 47(11):1279. Münzel, Kuntic, Daiber & Sørensen, Anatolian Journal of Cardiology 29(10):525. Tabaei et al., BMC Cardiovascular Disorders (2025). Schmidt et al., European Heart Journal 34:3508 (2013). Osborne et al., European Heart Journal 41:772 (2020). Saucy et al., European Heart Journal 42:835 (2021). Transportation noise and cardiovascular health review (2025), PMC12503095. Journal of Urban Health dose-response meta-analysis (2023). German national noise burden study, PMC6651346; Hesse road traffic noise burden, PMC8431690.

Sleep. Basner & McGuire, IJERPH (2018). Smith, Cordoza & Basner, Environmental Health Perspectives 130(7) (2022). Haralabidis et al. (HYENA), European Heart Journal 29(5):658 (2008). Basner, Müller & Elmenhorst, SLEEP 34(1):11–23 (2011). Basner et al., SLEEP 49(5):zsag001 (2026). Halperin (2014). Riedy et al., Sleep Medicine Reviews (2021). Ding et al., Sleep Medicine (2025). Hugh, Wolter, Cushing et al., Pediatrics 133(4):677 (2014).

Low frequency and infrasound. Leventhall, assisted by Pelmear & Benton, review for DEFRA (2003), incorporating Watanabe & Møller (1990), Kjellberg et al. (1984), Persson & Björkman (1988), Persson et al. (1990), Landström et al. (1993). Moorhouse, Waddington & Adams, NANR45, University of Salford (2005; rev. 1, 2011). ISO 7196. Persson Waye et al., Life Sciences (2002). Baliatsas et al., Science of the Total Environment (2016). Michaud et al., Sleep (2016). NHMRC (2015). Liebich et al., Sleep (2022). Lechat et al., Journal of Sleep Research (2026). Poulsen et al., Environmental Health Perspectives (2019). Baumgart, Fritzsche & Marburg, JSV (2021), doi:10.1016/j.jsv.2021.116310; Pilger & Ceranna reply, JSV (2022), doi:10.1016/j.jsv.2021.116636; Holzheu & Hundhausen, JSV (2024), doi:10.1016/j.jsv.2023.117960. Salt & Hullar, Hearing Research (2010). Weichenberger et al., PLoS ONE (2017). Ascone et al., Scientific Reports (2021). Chapman & St George, ANZJPH (2013). Tonin, Brett & Colagiuri (2016).

Cognition and hearing. Stansfeld et al. (RANCH), The Lancet (2005). Clark et al., Journal of Environmental Psychology 76:101646 (2021). Kim & de Dear (2013). ISO 3382-3. Buxton et al., PNAS (2021). Alvarsson, Wiens & Nilsson, IJERPH (2010). Dillard et al., BMJ Global Health (2022). Ramakers et al., JAMA Otolaryngology (2016). NIOSH recommended exposure limit. NIDCD, Noise-Induced Hearing Loss.

Music and entrainment. Hole et al., The Lancet (2015). Bradt et al., Cochrane (2021), cancer. Bradt & Dileo, Cochrane (2014), critical care. Aalbers et al., Cochrane (2017), depression. van der Steen et al., Cochrane CD003477.pub5 (2025), dementia. MIDDEL (2025) and the MIDDEL Australia cohort (2022). LongSTEP, PMC10220519. Yoo & Kim, Journal of Music Therapy (2016). Ghai & Ghai, Scientific Reports (2019). Koshimori et al., Annals of the New York Academy of Sciences (2025). Loomba et al. blood-pressure meta-analysis; BMC Cardiovascular Disorders (2016). Journal of Clinical Medicine 11(4):991 (2022). Salimpoor et al., Nature Neuroscience (2011). Mallik et al. (2017). Stanhope & Weinstein, Complementary Therapies in Medicine (2020). Ingendoh, Posny & Heine, PLOS ONE (2023), PMC10198548. Garcia-Argibay, Santed & Reales, Psychological Research (2019). Basu & Banerjee, Psychological Research 87 (2023). Xiong et al., Complementary Therapies in Medicine (2025). Shukla et al., International Journal of Clinical Pediatric Dentistry (2025), PMC12451561. Schwarz & Taylor, Clinical Neurophysiology (2005). Frontiers in Human Neuroscience (2017), PMC5694826. Geriatrics & Gerontology International 24(3):297–304 (2024). Da et al. (2024), PMC11522122. Chan et al., Alzheimer's & Dementia (2025), PMC12552893. Oster, Scientific American (1973). Fang et al. (2021).

Schumann and EMF. Schumann (1952); Balser & Wagner (1960–63). ICNIRP guidelines (2020) and nerve-stimulation restrictions (2010). Rubin et al., Bioelectromagnetics 31:1–11 (2010), PubMed 19681059. Karipidis et al., Environment International 191:108983 (2024) — Part I; Part II (2025). Melnick, Moskowitz, Hardell et al. (ICBE-EMF), Environmental Health (2025). IARC Monographs advisory group, The Lancet Oncology (April 2024). Ahlbom et al., British Journal of Cancer (2000). Amoon, Swanson, Magnani, Johansen & Kheifets, "Pooled analysis of recent studies of magnetic fields and childhood leukemia," Environmental Research 204 (2022), PMID 34481821, DOI 10.1016/j.envres.2021.111993 — added in the second corrections round. SCHEER, Final Opinion on potential health effects of exposure to electromagnetic fields (EMF): Update with regard to frequencies between 1 Hz and 100 kHz, adopted 17 May 2024 (European Commission, scheer_o_063) — read directly from the PDF; the adoption date and the "weak evidence" finding are quoted verbatim from it. SCENIHR (2015) — superseded by the above, and retained only as the earlier position. Cherry, Natural Hazards (2002). Deming, Journal of Scientific Exploration (2004).

Pitch standards and their paperwork. Arrêté of the Minister of State, 16 February 1859 (diapason normal, 870 simple vibrations at 15 °C), following the commission of 17 July 1858. International pitch conference, Vienna, 16–19 November 1885. Treaty of Versailles, Part X, Article 282, item (22) — verified at Wikisource. G. W. C. Kaye, "International Standard of Concert Pitch," Nature 143:905–906 (27 May 1939). Lindley et al., "Pitch," The New Grove vol. 14 (Macmillan, 1980), 785. Lynn Cavanagh, A brief history of the establishment of international standard pitch a=440 hertz, drawing on Ll. S. Lloyd, JRSA 98 (16 December 1949). ISO Recommendation R 16 (1955); ISO 16:1975, Acoustics — Standard tuning frequency, ISO/TC 43, catalogue csnumber 3601 (read via the iteh.ai mirror; iso.org returned HTTP 403). Legge 3 maggio 1989, n. 170, Gazzetta Ufficiale Serie Generale n. 109 (12 May 1989). Laurent Rosenfeld, "How the Nazis Ruined Musical Tuning," Executive Intelligence Review 15:35 (2 September 1988), 54. Rosenberg, Journal of Popular Music Studies 33:1 (2021), 137–154. Fanny Gribenski, Tuning the World: The Rise of 440 Hertz in Music, Science, and Politics, 1859–1955 (Chicago, 2023). Schiller Institute archive. SFU Sonic Studio handbook.

Devices and digital signal processing. Apple, Logic Pro tuning documentation. Ableton Live 12 manual, .scl / .ascl tuning and MPE routing. Rubber Band Library integration and technical documentation; the developer's engine-performance note (thebreakfastpost.com, 30 September 2022). Equalizer APO ticket #9, "Pitch support (shift of frequencies)" (opened 4 August 2015). Korg CA-50 and DT-4 manuals. Yamaha P-45 and P-71 owner's manuals. ffmpeg asetrate / aresample / atempo filters.

Continuous emissions and the Earth's own signals. ICNIRP low-frequency guidelines (2010). WHO Fact Sheet 322 (2007). IARC Monographs vol. 80 (2002). Tracking SDG7: The Energy Progress Report (2025). ZEVS; Project Sanguine / Project ELF; VLF Transmitter Cutler. Fraser-Smith et al., Stanford — 82 Hz reception at antipodal distances. NIST, WWV/WWVH broadcast format and WWVB coverage. IS-GPS-200. MIL-STD-704. Bell Precise Tone Plan. NERC, time error correction. NEMA ST-20 and TR-1; IEEE C57.12.90 (test code). Rhie & Romanowicz, Nature 431:552–556 (2004). Nishida, Annu. Rev. Earth Planet. Sci. 41:719–740 (2013) and GJI 199:416 (2014). Zhou, Yu, Cao & Qiao, JASTP 98:86–96 (2013). De et al., JASTP 72 (2010). Jacobs et al. (1964); USGS OF 2013-1166. Yagova, Pilipenko, Sakharov et al., Earth Planets Space 73:88 (2021). Longuet-Higgins (1950); Hasselmann (1963); Traer & Gerstoft, JGR (2012). ISO 2631-1:1997. Saroka, Vares & Persinger, PLOS ONE 11:e0146595 (2016).

Companion study: 432 Hz — the frequency that was never taken, slug 432-hz in studies.db.