Guess the Song

What is in a tenth of a second

Almost always: a drum. Popular music puts something percussive on the downbeat, and the excerpt Apple serves is a 30-second cut taken from somewhere in the middle of the track rather than the top, so the clip tends to land on a bar line rather than on an intro. A kick reads as a soft thud with no top end. A snare is a burst of white noise with a pitched crack under it. A closed hi-hat is all top end and no body.

That single hit rules out more than it names. A drum machine has no variation in the attack and no room tone around it, which places you in the 1980s or in electronic music. A live kit brings the sound of the room with it, and the length of that tail tells you whether the record was made in a big studio in the 1970s or in a treated box in the 2010s. Tape hiss, if you can hear any at all in 100 milliseconds, is its own answer.

Occasionally the clip opens on a voice, which is a gift: the front of a sung syllable carries a lot of identity, and most people can put a name to a singer from one vowel.

The ladder

0.1 seconds, then 0.5, 1.5, 3 and 5. The jump that changes the round is the third one. At half a second you have a sound; at a second and a half you have two or three beats, which means you have a tempo and a groove, and tempo alone eliminates most of a shortlist.

Three seconds is usually a full phrase, and five seconds nearly always brings the vocal in. Getting there costs you 850 of the 1000 points the round opened with, so five seconds is a confession rather than a strategy.

Audio is unlocked once and then each new rung plays itself, so buying more does not mean pressing play again. Space replays whatever you have paid for.

Where the music comes from

The deck is built from 178 tracks by 114 artists, played through the iTunes Search API, which serves preview clips to anyone without a key or an account. Nothing is stored or redistributed — the repository holds titles and artist names, and the audio is fetched at the moment you play.

That API indexes the store, not the history of recorded music, and the store is full of note-for-note impostors. A search for a well-known rock song will happily return a lullaby version, a string-quartet tribute, an 8-bit cover and a karaoke backing track ahead of the real thing. Anything matching that pattern is discarded before it can become a round, which is why you never get half a second of a music-box arrangement and have to guess what it is a cover of.

Clips play through Web Audio rather than an audio element, because an audio element cannot honour 100 milliseconds. Seeking and pausing both jitter by tens of milliseconds, which on this ladder is most of the clip. Buffers are decoded up front and scheduled sample-accurately, with about 12 milliseconds of fade on each end to kill the click.

Scoring a run

Ten rounds. A first-rung answer is worth 1000, then 620, 380, 240 and 150 as you buy your way down. A wrong guess costs a rung exactly like a skip does, so a stab in the dark at 0.1 seconds is a real bet: guess wrong twice and you are at three seconds with 240 points on the table.

The speed bonus adds up to 160 points and tapers to nothing over 20 seconds, and the streak multiplier climbs by a tenth per consecutive correct round to a maximum of 2×. A run where you name everything from the first tenth of a second inside ten seconds is worth a shade under 16,000. Anything over 6,000 means you were mostly winning on the first two rungs.