Somewhere in a product you use today, five short tones are doing five different jobs — a boot, a refresh, an error, a success, a notice. Tap each one below and guess which is which, before you're told anything at all.
Scroll to hear them
—
—
—
—
—
Guessed
0/5
Tap a sound, then tap the label you think it is.
Sound is off — flip the Sound switch in the Signature Bench below to actually hear these. Until then the waveform above is your only clue, which is the point: this page follows its own subject’s Rule 5 — sound is never the only channel.
Part I
The Signature Bench
Every sound above was built from the same handful of levers: how many notes, how far apart they sit, how long each one rings, how fast it arrives, and what it's made of.
Below is the bench those levers live on. Build a two-to-four-note motif and watch the waveform and the piano roll answer instantly — sound stays off until you flip the switch, and even then, only the speaker button plays anything. Try the two presets first.
The Signature Bench
build a 2–4 note motif
Duration
660ms
Pitch range
392Hz–1109Hz
18 semitones
Reads as
Neutral
Modelled — formula below.
ConsonantDissonant
SnappyLong
InstantSoft
Cut offTail
Timbre
Duration and pitch range above are computed straight from the parameters that drive playback — exact, not modelled. “Reads as” is different: warmthUrgency = 0.65×(consonance, rescaled −1..1) + 0.35×(attack, rescaled −1..1), clipped to [−1, 1]. It depends on nothing but interval and attack, on purpose — that is the whole demonstration the two preset buttons are built to show.
Predict
Set the Signature Bench to 3 notes. Can the identical note count read as either a warm welcome or a hard alarm, without changing anything else?
Part II
A Constraint, Not a Genre
Brian Eno joined Roxy Music as a synthesizer player in 1971, left in 1973, and spent the rest of the seventies moving from performance toward a compositional idea: music that didn’t demand attention but rewarded it if given.
In February 1979 he released Ambient 1: Music for Airports, reportedly born of hours spent irritated by the piped-in atmosphere at Cologne Bonn Airport — music too polite to notice and too present to ignore, doing neither job well. His own liner notes state the principle a good notification sound still has to pass on its thousandth playback:
Ambient Music must be able to accommodate many levels of listening attention without enforcing one in particular; it must be as ignorable as it is interesting.
Brian Eno, liner notes to Ambient 1: Music for Airports, 1979
Fifteen years later, in 1994, two Microsoft designers handed Eno a brief with no music in it at all — just adjectives: “inspiring, universal, blah-blah, optimistic, futuristic, sentimental, emotional.” The one hard number was duration: roughly three and a quarter seconds. Eno wrote eighty-four candidate pieces against that ceiling before delivering the one that shipped. On his own account, on BBC Radio 4’s The Museum of Curiosity, he composed it on a Macintosh — “I’ve never used a PC in my life; I don’t like them” — a private irony, not a stated act of sabotage.
The sound that actually shipped with Windows 95 runs about six seconds, not 3¼. The dossier behind this page flags the two figures as unreconciled rather than smoothing them into agreement — both numbers are given here for that reason, not because either has been quietly corrected to match the other.
Read as folklore, this is a story about a vague brief producing great work. Read precisely, it is the opposite: the adjectives were loose, but the duration ceiling and the discipline of generating and discarding eighty-three other attempts were not. A hard technical constraint doesn’t limit creative range — it forces selection to happen before delivery instead of after.
Two vocabularies exist for what plays inside a piece of software, and they solve different problems. An earcon — the term researcher D. A. Sumikawa coined in 1985, punning deliberately on “icon” — is a brief, abstract, synthesized tone standing in for a state: fast to process, faster than speech, and the reason a screen reader teaches its cues early so they become habitual. An auditory icon instead uses a recorded, real-world sound — a shutter click, a paper crumple — that resembles what it means, which trades brand-consistency for instant, untrained recognition. Confusing the two is a category mistake: use an earcon where you need a coherent brand family, an auditory icon where you need a stranger to understand it cold.
One family, not five effects
A sonic system is built once, then reused
A serious sonic identity is rarely one iconic sound. It is a small family — one anchor pitch or interval, reused across every state — so the set reads as one brand even heard piecemeal. Walter Werzowa’s 1994 five-note “Intel bong” has been re-orchestrated for three decades, including a 2015 arrangement layered against Beethoven’s Fifth for a Super Bowl campaign, and the underlying identity never stopped being recognizable. Netflix’s two-note “ta-dum” has run, unchanged, at a fixed point in the experience since 2014.
Apple’s startup chime and the NBC chimes make the same point from opposite directions. Jim Reekes built the Mac “bong” from a modified synth preset without formal sign-off, refused to revert it once discovered, and it became a registered U.S. trademark in 2012 — iconic because it was informal. NBC’s three-note chime began in 1926 as an engineering cue to synchronize station identification, was standardized by the 1930s, and received the first purely audible U.S. service-mark protection ever issued, in 1950 — durable because it solved a real problem before anyone called it branding. The popular retelling that credits Eno’s Windows 95 commission with inventing sonic branding is contested and imprecise for exactly this reason: NBC beat it by nearly seventy years.
Which leaves the question this whole page is built to answer: when is silence the right sound? Whenever a control already shows its own state change — a button that visibly depresses, a label that visibly updates — adding a chime on top confirms nothing new. It just trains a user to reach for the mute switch, and once an app is muted for one unbearable sound, every well-designed sound in it goes silent too.
Part III
Five Roles, Five Budgets
Duration is not a taste question here — it is a published ceiling per role, treated as a shipped defect if crossed.
Scrub across the five roles below. Same synthesis engine as the bench above, one role at a time.
Sound roles, by budget
scrub a role
A once-per-session milestone can spend the whole 3.25s ceiling Eno was given — it never repeats inside a task, so length stops being a tax (Rule 1, "when to break it").
Role
Boot
Duration budget
3250ms
Sound role
BootRefresh
Boot
Part IV
Every Event Beeps, or Only Meaning Does
A feature phone chimes at a keypress, a menu move, a low battery, an incoming text — a dozen distinct sounds sharing no motif, cluttered by the third one.
A modern OS spends its sound budget only at state changes a user genuinely cannot otherwise verify — send, receive, fail — and stays silent through everything else. Drag to compare the same screen under each philosophy.
⇆
The Nokia Tune traces to Francisco Tárrega’s 1902 “Gran Vals” and was reportedly heard an estimated 1.8 billion times a day by 2009 — a single, widely repeated estimate, not a re-verified count, but the direction of the claim (a short public-domain excerpt can carry a brand as durably as a bespoke one) is uncontested. The widely repeated claim that camera-shutter sounds are legally mandated and non-mutable in specific countries could not be confirmed against a primary source in this dossier and is flagged contested — the anti-pattern it illustrates (a decorative sound the user cannot mute) is real regardless of that specific legal claim.
Part V
The Cost of Sounding Too Much
Habituation forgives a sound heard fifty times a day, as long as it is short, simple, and consistent. It stops forgiving once too much of the app makes noise.
Raise the share of interactions that trigger any sound and watch a modelled annoyance curve answer.
Over-sounding
raise the share of interactions that get a sound
Modelled annoyance holds near 5 through light use, then accelerates past a habituation-breakdown point to 86 once every interaction sounds.
Sounded
20%
Modelled annoyance
9
Modelled — formula below, not a measurement.
RareEvery tap
Modelled from the psychology in Part 6 of the dossier — the mere-exposure effect’s diminishing, then negative, returns, and the notification-spam anti-pattern in Part 9 — not measured against real users. The exact curve is illustrative; the shape (slow, then an accelerating breakdown) is the point.
Challenge · One app, two sounds
Open
Design the success and error sounds for one app so a blindfolded user can tell them apart by ear — and keep both under 400ms.
Success
440ms
ConsonantDissonant
Snappy400ms
Error
440ms
ConsonantDissonant
Snappy400ms
Success is still missing something: pull its Interval slider toward Consonant, and keep it at 400ms or under.
Insight
A brand you can hear is a brand you cannot look away from.
A logo waits for the eye to land on it. A sound trademark installs itself at the exact instant something happens — a send, a save, a failure — whether or not anyone was looking at the screen. That is precisely why the discipline in this manual is defensive rather than expressive: the same channel that can make a product recognizable with the eyes closed is the one channel a user cannot simply choose not to look at. Eno’s real contribution wasn’t a genre. It was proving that the fix for that power is a constraint, applied before delivery — not restraint, applied after the fact.
Mental model
Whenever you are about to add a sound to an interaction,
look for the invisible state it confirms, its role's duration budget, and whether it shares your system's one anchor interval,
because a sound with no state to confirm is a tax on attention, and a sound outside the family breaks the brand it was meant to build.
Tomorrowloop your loudest confirmation sound for two minutes straight. If it's still tolerable at minute two, ship it. If it isn't, it never should have been.