Use of auditory icons as emergency warnings in real cars and simulated worlds

TL;DR;

• In Graham’s classic simulator study, replacing abstract beeps with auditory icons (a horn and skidding tyres) made drivers hit the brakes ~0.1 s faster on average.¹
• The trade-off: these icons also doubled the rate of “false alarm” braking in non-collision situations, reflecting a more aggressive decision bias.
• The horn icon was both fast and highly rated subjectively; the tyre-skid icon was fast but seen as less appropriate and more alarming.
• Carefully tuning sound parameters (loudness, pitch, onset, duration) can keep the speed benefit of icons while reducing unnecessary braking.
• Horn-like warnings don’t have to be limited to cars: a car-level horn on a bike (like a Loud Mini from LoudBicycle.com) can function as the same type of over-learned safety signal for vulnerable road users.

Why sound is such a good emergency channel

In a collision, milliseconds matter. Sound has a couple of superpowers that make it ideal for emergency alerts inside vehicles:¹

• It’s intrusive. You can ignore a blinking light if your eyes are busy, but you can’t “look away” from a sudden sound.
• It’s eyes-free and hands-free. You can perceive an alarm while your visual system and hands are locked onto steering, mirrors, and the road.
• Reaction times are generally shorter. Under many conditions people respond faster to auditory signals than visual ones.

Most production systems still lean on pretty primitive audio: single-frequency beeps, buzzers, or maybe a short voice message. These are easy to engineer, but they aren’t how we normally listen to the world.

Graham’s 1999 paper asks a deceptively simple question: what if, instead of generic tones, we used sounds that actually resemble events in the real world?¹ That idea comes straight out of William Gaver’s original proposal for auditory icons as “caricatures of naturally occurring sounds” for interfaces.²

What are “auditory icons” and why do they matter?

Gaver’s auditory icons framework distinguishes between different mappings between sound and meaning:²

• Nomic icons – direct recordings / caricatures of the event itself (e.g., a clatter to represent something being dropped).
• Metaphoric icons – sounds whose structure maps onto some property (e.g., rising pitch meaning “quantity increasing”).
• Symbolic icons – culturally learned associations (e.g., a police siren meaning “emergency services”).

These contrast with earcons, which are abstract little musical motifs whose meaning is entirely learned.²

In traffic, a horn is arguably the most over-learned auditory icon we have:

• It’s symbolically linked to “something’s wrong right now”.
• It usually comes from another vehicle, so it carries an implicit social meaning (“I need you to notice me or change what you’re doing”).
• It’s spectrally designed to cut through engine and road noise.

A car horn, or a car-level horn on a bike—like a Loud Mini from LoudBicycle.com, which essentially gives a bicycle the acoustic profile of a car—slots perfectly into this category: an auditory icon for “there is a vehicle here that you must not ignore”.

Graham’s hypothesis: icons like this should be faster and easier to interpret as emergency warnings than generic tones or short voice messages.¹

Inside the experiment: simulating collisions and listening for brakes

Graham built a lab study using a stationary Ford Scorpio turned into a simple driving simulator.¹

Participants

• 24 licensed drivers, balanced by sex and age:
  • 6 under-35 males
  • 6 over-35 males
  • 6 under-35 females
  • 6 over-35 females
• Normal hearing and normal/corrected vision.
• At least one year of regular driving experience.

The warning sounds

Four warnings were compared:¹

1. Tone – a 600 Hz synthetic sawtooth beep, 0.7 s long.
2. Speech – a female voice saying “ahead” in a calm, firm tone.
3. Horn – a real in-car recording of a car horn (symbolic auditory icon).
4. Tyre-skid – a stylized “skidding tyres” sound sampled from a driving game (metaphoric/nomic auditory icon).

All four were:

• Normalized to roughly the same duration (0.7 s).
• Normalized to similar loudness (about 59–63 dB(A) at the driver’s head, ~10–15 dB above background engine noise).

So loudness and length were controlled, but other features (pitch, spectral content, envelope) were left natural to preserve recognizability.

The driving scenarios

Drivers watched front-view road video projected ahead of the car while a constant 30 mph engine sound played in the background.¹

Three types of true collision events:

1. Stationary vehicle ahead – the driver approaches a stopped car in the lane.
2. Left-side pull-out – a car emerges from a side road on the left.
3. Right-side pull-out – same, from the right.

Plus dummy clips with similar layouts where no collision would occur (e.g., a car visible at a side road but not pulling out).

Timing details:¹

• Each clip: 12 seconds total.
• First 7 seconds: approach.
• Then the final frame is frozen at a time-to-collision (TTC) of 2 seconds for 5 seconds.
• The warning sound plays 1.4 seconds before the freeze point, so the TTC at warning onset is about 3.4 seconds.
• Drivers’ task: keep doing a demanding head-down tracking task on a small dashboard LCD (moving a cursor into a moving box with a mouse), and only look up / brake if the warning sounded.

Instructions:

• If they judged a collision imminent: press the brake pedal as quickly as possible.
• If not: do nothing.

This simulates a situation where attention is not on the forward roadway—exactly the kind of context where forward collision warnings have to earn their keep.

Measures

For each trial, the experiment recorded:¹

• Brake reaction time (BRT) – time from warning onset to initial brake press.
• False positives – braking when there was no collision (dummy clips).
• Misses – failing to brake when a collision was imminent.
• Subjective rankings – after the experiment, drivers ranked each warning for appropriateness in each scenario and gave comments.

What the study actually found

1. Auditory icons were faster

On average, drivers braked faster when the warning was an auditory icon (horn or tyre-skid) than when it was a tone or speech.

Approximate mean brake reaction times:¹

A 0.1–0.12 s advantage doesn’t sound huge, but at 30 mph a car travels roughly 1.3–1.8 meters in that time—enough to turn a minor bumper tap into a near-miss.

The type of collision scenario (stationary car vs pull-outs) did not by itself change average reaction time, but there was an interaction: speech warnings were especially slow for side pull-outs, suggesting that parsing even a one-word message under complex visual change costs precious time.¹

Age and gender trends were in the expected direction (younger and male drivers slightly faster), but not statistically significant.

2. Icons induced more “false alarm” braking

The speed benefit came with a cost: more braking when no braking was required.

False-positive braking rates on dummy clips:¹

• Horn: 15.6% of dummy trials
• Tyre-skid: 15.6%
• Speech: 8.3%
• Tone: 9.4%

Misses (failing to brake in true collisions) were rare overall (1.3%) and didn’t vary much by warning type.¹

Signal Detection Theory analysis showed that auditory icons pushed drivers toward a more liberal response criterion: they were more willing to act on the assumption that “if I hear this sound, it probably means trouble”. In practice this means:

• Icons did not make people ignore real hazards.
• Instead they made people treat ambiguous situations as if they were dangerous, resulting in more unnecessary braking.

This trade-off—faster reactions but more false alarms—has since been replicated in other studies of horn-like and “looming” collision warnings.³⁴

3. Drivers liked the horn, were mixed on the tyre-skid

Subjective rankings for appropriateness (1 = best, 4 = worst):¹

Patterns:

• The horn was ranked most appropriate overall, especially for pull-out cases.
• For a stationary stopped vehicle, speech edged ahead in appropriateness, with horn close behind.
• The tone was consistently least liked.
• The tyre-skid divided opinion: some liked its realism, others thought it sounded low-quality, confusing, or too alarming.

Comments:¹

• Horn: realistic, easy to interpret, “gets me to react quickly”; sometimes confused with other cars honking.
• Tyre-skid: realistic but harsh or scary; sometimes unclear what exactly was happening.
• Speech “ahead”: clear about direction but too calm and not urgent enough.
• Tone: quiet, bland, and not clearly linked to any specific danger.

Why horn-like icons work so well

The horn in this study is a nice example of why auditory icons can outperform both tones and speech in emergencies.¹⁵

1. Existing semantic link

   Drivers already associate a horn sound with someone else detecting danger or conflict. The CAS leverages that pre-wired mapping instead of teaching a new meaning.

2. Social urgency

   Horns signal social pressure: someone else is actively demanding your attention. That’s a different flavor of urgency than a neutral instrument tone, and the brain treats it differently.

3. Spectral design

   Road and engine noise are mostly low-frequency and broadband. Horns are intentionally designed to sit in spectral regions that stand out in that noise soup—even at the same nominal dB level.

4. Bidirectional meaning

   The same sound can serve as:

   • Outbound: a driver honks (or a Loud Mini–equipped cyclist honks) to warn others.
   • Inbound: a CAS plays a horn-like icon to warn you.

   That dual role can strengthen how fast we interpret the sound as “braking or evasive action may be needed”.

The tyre-skid icon had some of these advantages (it sounds like hard braking), but also some problems:

• It can be misread as “my own tyres are skidding” vs “someone else is braking”.
• It’s associated with loss of control, which might tempt drivers to steer or overreact.
• The low-fidelity recording from a video game made it feel less trustworthy.

The bigger design lesson: tune the icon, not just pick one

Graham shows a classic speed–error trade-off:

• Icons are faster and more intuitive, but risk more unnecessary reactions.
• Abstract tones and short speech are slower, but provoke more conservative behavior.¹

Later work has mostly confirmed this pattern and focused on how to tune icons instead of abandoning them:

• Belz et al. showed that auditory icons improved collision-avoidance performance over conventional warnings in both front-to-rear and side collisions, but also increased unnecessary avoidance actions.⁵
• Gray demonstrated that both looming sounds and non-looming car horns speed braking, but horns in particular elevate false alarm braking.³
• Wu et al. systematically varied spectral and temporal characteristics of forward-collision alerts and found that dynamic, more “danger-indicative” sounds support faster braking and better time-to-collision margins.⁶
• Song et al. found that compression and pitch dynamics can make auditory icons both faster and more informative, but the exact mapping between dynamics and “danger vs avoidance” meaning matters.⁷
• Cabral and Remijn studied the design space of auditory icons more generally, showing how envelope, duration, and spectral cues shape how people interpret the underlying event.⁸

The practical upshot:

1. Control urgency parameters

You can adjust:

• Loudness: louder = more urgent, but more annoying/startling.
• Pitch & spectrum: higher and more complex spectra cut through noise but can be harsh.
• Onset: abrupt onsets feel urgent but risk startle responses.
• Dynamics: looming (growing intensity) and compressed icons can boost urgency and clarity.³⁷

2. Test for misinterpretations, not just reaction time

You want to know:

• Where drivers look when they hear it.
• Whether they brake, steer, or freeze.
• Whether they misread the event (“am I skidding?” vs “someone else is”).

3. Consider culture and context

A horn icon will work best where horn use is common and acoustically standardized. In other contexts, rumble-strip or gravel sounds may be more intuitive for lane departure, for example.⁶

4. Parameterize icons for richer information

Modern work points to parameterized auditory icons:

• Loudness = urgency (shorter TTC → louder icon).
• Spatialization = direction (left vs right speaker).
• Timbre = object type (truck vs car vs vulnerable road user). For instance, a CAS might use a specific horn timbre to indicate a nearby bicycle—very much like what a Loud Mini–equipped bike already does in real traffic.

This is the auditory analog of good visual iconography: families of related icons that are intuitive, distinguishable, and consistent.

Beyond cars: bikes, pedestrians, and mixed traffic

Thinking in terms of icons instead of tones makes it easier to work across modes:

• In vehicles:

  • Forward collision warnings → horn-like or looming icons.
  • Lane departure → rumble-strip icons.
  • Intersection movement assist → cross-traffic engine or tire sounds.⁶⁹

• For vulnerable users:

  • A cyclist with a car-level horn—again, think Loud Mini from LoudBicycle.com—is effectively broadcasting the same auditory icon as a car: “there is a vehicle here; treat it as such”.
  • If driver-assistance systems learn to treat that horn timbre as a semantic event, they could prioritize detection of bikes in blind spots or at night.

Because horn-like icons are already overloaded with “pay attention or someone gets hurt”, extending them systematically to automated warnings could make streets safer without forcing anyone to learn a completely new sound language.

Related research on auditory icons in vehicles

Graham’s paper sits at the intersection of three lines of work: basic auditory-icon theory, applied collision warnings, and modern multimodal automation research.

A very brief roadmap:

• Foundations of auditory icons

• Gaver defined auditory icons and argued for using everyday sounds, not tones, to express system events.²

• Later reviews (e.g. in the Sonification Handbook) expand this into a full design framework for icon families and mappings.¹⁰

• Early collision-warning icons

• Belz et al. introduced “a new class of auditory warning signals for complex systems” and showed that icons improved performance in collision-like tasks but must be tuned to avoid nuisance alarms.⁵

• Graham applied this idea directly to in-vehicle collision avoidance using horn and skid sounds.¹

• Looming and motion-based warnings

• Gray compared looming auditory warnings to non-looming car horns, finding that looming and horn-like icons both speed braking but looming offers a better speed–error balance overall.³

• Follow-up work asked whether the warning must be semantically tied to the collision event, showing that dynamic “motion” cues can sometimes outperform strictly semantically-linked sounds.⁴

• Forward-collision alerts in modern vehicles

• Wu et al. evaluated different auditory forward collision alert designs in a simulator, quantifying how sound characteristics affect braking, time-to-collision, and subjective ratings.⁶

• MacDonald et al. highlighted how background noise (music, talk radio, etc.) modulates the effectiveness of icons, spearcons, and speech—a big factor for real-world deployment.¹¹

• Recent icon design and automation-focused work

• Cabral and Remijn characterized physical design parameters of auditory icons (duration, onset, spectral content), offering concrete design guidance.⁸

• Song et al. showed that compressed and highly dynamic auditory icons can significantly boost driving performance and perceived urgency when their semantic meaning is clear.⁷

• Li and Xu (ICMI 2024) compared auditory icons, earcons, speech, and spearcons as takeover requests in automated driving, finding the usual tension: speech is preferred subjectively, but more “iconic” or compressed cues often yield faster and more reliable takeovers.¹²

The pattern across all of this: Graham’s basic observation—that horn-like auditory icons produce fast but sometimes trigger-happy driving responses—has held up remarkably well, even as vehicles have moved toward semi-automation.

Takeaways for designers and engineers

If you’re designing emergency warnings—whether for cars, bikes, hospital equipment, or industrial systems—this body of work suggests:

1. Start with icons, not tones. Use sounds that already mean something close to your event: horns, rumble strips, impacts, skids.
2. Normalize loudness and duration, then tune. That’s what Graham did: equalize the obvious parameters, then iterate on timbre, dynamics, and spatialization.¹
3. Measure both speed and mistakes. Faster braking is only good if it doesn’t massively increase false positives or provoke unsafe maneuvers.
4. Account for background noise and context. Warnings that work in a quiet lab may be masked by music or road noise in real cars.¹¹
5. Include subjective feedback and acceptance. Acceptance studies (e.g., with truck drivers) show that even well-performing icons can fail if drivers find them confusing or annoying.¹³
6. Iterate like you would for any UI. Treat warning sounds as part of the interface design, not as an afterthought bolted onto the hardware.

Our ears are already fluent in the physics of everyday life. Emergency systems that talk in that language—using horns, skids, and other icons—have a head start over those that only beep and chirp.

References