Auditory icon

William Gaver proposed these in 1986, as the deliberate alternative to a screen full of beeps. His argument was that everyday listening is not listening to sound, it is listening to events: nobody hears a spectrum, they hear a heavy thing dropped on wood. So the sounds an interface makes should be caricatures of naturally occurring sounds, exaggerated recordings of real events, and the mapping to what happened should be as close to physical as the interface can manage. He sorted the possibilities into three kinds: symbolic, where the relation is pure convention, metaphorical, where it is a likeness, and nomic, where the sound genuinely is what the thing would sound like. His SonicFinder made the case in a real interface, where dragging a file scraped, dropping it thudded, and a big file sounded bigger than a small one.

That is the whole discrimination against its sibling. An earcon is a synthesized figure whose meaning is learned, so a set of them has to be taught and can be built into a family; an auditory icon is a recording whose meaning is recognised, so it needs no teaching and cannot be extended by transposing anything. The trade runs in both directions. Nobody has to be told that a crumple means the thing is gone, but the world only supplies sounds for events that happen in it, and there is no everyday sound for “your two-factor code expired”. Interfaces in practice mix the two registers, which is fine as long as they do not mix them for the same family of events.

Being a metaphor delivered as sound puts the term next to the visual metaphors and next to their failure modes. It is the audio branch of an interface metaphor, and it goes wrong the same way skeuomorphism does: the reference ages out. A camera shutter on a phone with no shutter, a modem handshake, a typewriter carriage on a software keyboard, all of them were once nomic and are now a costume, understood by people old enough to have heard the original and mystifying to everyone else. The rule of thumb is to borrow from physics rather than from equipment, since paper still crumples and heavy things still land, while machines get replaced.

Then there are the duties every sound in an interface owes, and they are the reason the specimen here is silent until you press it yourself. Nothing plays unasked, the sound is never the only report of anything (a muted phone is the normal case, and a person who cannot hear it is owed the same information a listener gets), and a recording is heavier than a tone in every sense: more bytes, more room to sound cheap, and much more likely to become grating on the fortieth repetition. Where the noise is standing in for haptic feedback or for a visible change, ship the visible change too and treat the sound as the part that can be turned off.