EVs, loud in all the wrong ways

Both sides of this argument are right, and both are missing the point

It's a familiar exchange by now, playing out somewhere on social media most weeks. Someone posts that EVs should make noise, because silent cars are a hazard for blind pedestrians. Someone else replies that EVs already do make noise — it's been required by law for years. Both are correct. Neither has actually engaged with the only question that matters: what sort of noise.

The first poster has a genuinely good instinct and an outdated picture — EVs have made noise below 20-30 km/h for years now, under a mandate that predates most of these social posts by half a decade. The second poster is technically right and argues past the point entirely: the fact that a sound exists says nothing about whether it does the one job a warning sound has to do, which is tell a pedestrian where the car is. Two people, two true statements, and the actual engineering question sitting untouched in the middle of both of them.

That question is worth taking seriously, because on closer inspection, the sound regulators mandated gets it wrong twice over — once on where it's loudest, and once on whether anyone can tell where it's coming from.

Electric cars really are dangerously quiet

Start with the premise both sides of that exchange agree on, because it's correct and worth stating precisely, rather than picking apart the fix, before going further. Below about 20 km/h in the EU or 30 km/h (18.6 mph) in the US, an electric car makes almost no sound — no engine, no exhaust, tires barely rolling.

The evidence for that being a genuine hazard, not just an intuition, predates the modern AVAS rules. A 2009 NHTSA study found hybrid-electric vehicles were twice as likely as combustion vehicles to be involved in a pedestrian crash during low-speed maneuvers specifically — slowing, stopping, reversing, parking — the exact conditions where a hybrid switches to its silent electric motor. That finding is what triggered the 2010 Pedestrian Safety Enhancement Act and, eventually, FMVSS 141 in the US; the EU's AVAS rule followed a separate but similarly motivated process a few years later.

That's not the whole picture, though — as a December 2025 University of Leeds analysis of UK collision data makes clear. The study — casualty counts spanning 2014-2023, casualty rates calculated over the more recent 2019-2023 window — found that fully electric vehicles, across the full range of driving conditions, aren't associated with a higher pedestrian casualty rate than combustion cars at all. Hybrids still show a markedly higher rate — but the researchers attribute at least part of that to hybrids' outsized use as taxis and private-hire vehicles racking up disproportionate urban mileage, not necessarily to the acoustic mechanism alone.

None of that 2025 nuance was available when the mandate was actually written, of course — regulators acted years earlier, on the 2009 finding alone. Below 20 km/h in the EU or 30 km/h in the US, EVs and hybrids now have to emit an Acoustic Vehicle Alerting System (AVAS) tone, switching off once tire and wind noise take over at higher speed. The EU standard sets overall sound levels between 56 and 75 dB(A) depending on vehicle type; the US standard instead sets minimum levels across specific frequency bands rather than one overall figure, but the audible range lands in the same territory. Problem identified, fix mandated. Case closed.

Except it isn't.

The fix is loudest exactly where it's needed least

Regulators set AVAS loudness requirements with typical urban ambient noise in mind — audible over traffic, at a level comparable to a combustion engine — not tailored to whatever street the car actually happens to be on. That makes sense for a busy street. It makes considerably less sense for the quiet residential street where an EV owner is pulling out of their own driveway at 4 km/h.

On that street, ambient noise is low to begin with — which is precisely why pedestrians there don't need much help hearing a car coming. And it's precisely where a sound set for urban ambient becomes the loudest thing around. The AVAS mandate solves a real problem in the environment where the problem barely exists, and creates a new one — unwanted noise — in the environment that needed the least help.

That's irony one, and it's arguably the smaller of the two.

There's a parallel worth a beat here. Airport noise regulation is having almost the identical argument right now. The FAA's default metric, DNL, is a flat 65 dB threshold — an absolute number, applied the same way whether the airport sits in a dense city or a quiet valley. Community groups pushing back on that, in the FAA's ongoing Noise Policy Review, have proposed the opposite: a relative metric — often written N-Above-Ambient+10 — that counts events a fixed number of decibels above whatever the local ambient actually is, precisely because a flat threshold can't tell the difference between a loud sound in a loud place and the same loud sound in a quiet one. AVAS has the identical structural flaw at a much smaller scale: a fixed minimum sound level, applied the same way to a six-lane arterial and a cul-de-sac. The fix, in principle, is the one aviation-noise advocates are already asking regulators for elsewhere — set the requirement relative to ambient, not as a flat number chosen with an average street in mind.

A two-tone whine is nearly impossible to place

Irony two is worse, because it undermines the safety case itself, not just the noise nuisance case.

Most AVAS implementations use a continuous tone, frequently two-tone. A 2025 study by Leon Müller, Jens Forssén, and Wolfgang Kropp — a 24-loudspeaker listening experiment with 52 participants, published in the Journal of the Acoustical Society of America — tested how well people could localize combustion-engine noise against two-tone AVAS, multi-tone AVAS, and narrowband-noise AVAS. Two-tone AVAS came last, and not narrowly. Twenty of the 52 participants failed to correctly localize more than half the vehicles under two-tone AVAS. Under combustion engine noise, only 7 participants had any localization failures at all. Errors ran up to 27 degrees off, response times over 4 seconds slower.

The researchers' own explanation is straightforward psychoacoustics: broadband sound — like an engine — carries information across many frequencies, and the auditory system uses differences across that spectrum to work out direction. A narrowband tone doesn't give it much to work with. Their recommendation was to introduce randomness, so identical vehicle models stop emitting identical, identically-hard-to-place tones. A follow-up study by the same group (January 2026) found the same two-tone signals also scored worst on physiological stress and annoyance, among the outcomes it measured, when heard indoors.

So the sound regulators mandated to make EVs noticeable is, by design, one of the harder sound types for the human ear to place. Pedestrians can tell something's there. They're measurably worse at telling where — which is exactly the gap in the social-media exchange above. The second poster is right that EVs make noise. They're not right that the noise helps a blind pedestrian work out which direction to step away from, or how many vehicles are involved. "Making noise" and "making the right noise" aren't the same claim, and this is the study that shows the difference isn't academic.

Sharp attacks beat steady tones — but that specific claim hasn't been tested yet

Here's where it's worth being precise about what's evidence and what's a reasonable guess, because they don't carry the same weight.

The localization mechanism the Müller studies describe — broadband spectral content helping the ear triangulate — points toward an obvious alternative: a percussive or transient-onset sound, rather than a continuous tone. Sharp onsets and broadband content are exactly the features a steady two-tone whine lacks. It's a reasonable extrapolation from the same psychoacoustic mechanism the study identifies.

It isn't, however, something the study tested. Nobody in this literature ran a percussive AVAS design through the same 24-speaker rig and measured localization error against it. The claim that a percussive design would outperform a continuous tone is an informed inference, not a citable result — worth stating plainly rather than smuggling in as settled science. A related but distinct claim — that a non-steady sound can beat a steady one on plain detectability, not azimuthal precision — does have direct EV evidence, covered further down.

The playbook for this already existed

What makes the continuous-tone choice frustrating rather than just unlucky is that the field working out how to make warning sounds both attention-grabbing and placeable isn't new. It's decades old.

Roy Patterson wrote the UK Civil Aviation Authority's 1982 guidelines for auditory warning systems on civil aircraft, and in 1986 — with Edworthy, Shailer, Lower, and Wheeler — researched alarm sounds for intensive care and operating theatre equipment. That work underpins most of the auditory-warning design standards still in use. The core finding, in short: a warning needs a distinct temporal and melodic pattern, not just volume, to be quickly learned and correctly interpreted. Loudness alone doesn't make a sound useful — its structure does.

I met Roy several times at conferences, back when auditory warning design was closer to my own field than my hobby, and admired his work for its practical application. This is a significant reason why this particular failure has needled at me for years, well before I'd tracked down the Müller study confirming it.

The principle isn't just adjacent to EVs, either — it's been tested on them directly, and for over a decade. A 2014 study by Etienne Parizet, Wolfgang Ellermeier, and Ryan Robart, published in Applied Acoustics, tested nine synthesized warning sounds on 153 participants, including 53 blind listeners, measuring how fast and accurately each let someone tell which direction an approaching EV was coming from. The two best performers both used irregular amplitude modulation — a pulsing structure, not a steady tone — and matched a noticeably louder diesel car for detectability. They also happened to have the lowest sound pressure levels of the nine sounds tested. The best warning wasn't the loudest one — which means the loud-on-quiet-streets problem from earlier isn't the unavoidable cost of being heard.

AVAS regulation solved for loudness. It didn't solve for structure.

Badly executed, this becomes a self-inflicted wound

None of this is an argument against warning EVs make noise — the underlying hazard is real, at least in the low-speed maneuvers where it's been measured, and doing nothing was never the honest alternative. It's an argument that regulators picked a sound profile that's loud in the wrong places and hard to locate everywhere, when the acoustic engineering for doing this properly was already sitting in aviation and hospital-alarm literature before the first AVAS mandate was written.

That gap matters beyond acoustics. A resident on a quiet street who finds AVAS obnoxious isn't manufacturing a grievance — the sound really is set for an environment they don't live in, and it really is harder to place than the engine note it replaced. A genuine, repeatable annoyance doesn't go to waste — "EVs are noisy and annoying" is now a complaint anti-EV arguments didn't have to invent. A safety mandate that was entirely justified on its own terms has become, through nothing but sloppy sound design, free ammunition against the EV transition it was never meant to threaten.

One design would answer both problems

The Müller studies' own fix — randomized, rather than identical, tones per model — is a step in the right direction. But there's a cleaner argument sitting underneath both problems at once. A continuous tone can't tell a pedestrian where the car is. It can't tell the car how loud to be, either — not without measuring an environment it's already drowning out. A pulsed, transient-onset design offers a way out of both — the same fix, arrived at by two completely different routes. Nobody's built and tested it yet. Somebody should.

Written in collaboration with Claude (Anthropic).