Bouncing balls that
play the song

The melody is held as an ordered queue of notes. When a collision occurs that qualifies as a trigger, the queue advances by one and that note is played. There is no timeline and no beat grid, which is why the result sounds tied to the motion rather than laid over it.

The consequence is that the physics is your sequencer. If collisions come faster, the song plays faster. If the run settles and impacts become sparse, the melody stretches out. This is a feature when the run is tuned for it and a problem when it is not, which is why setting gravity and arena size before attaching audio saves a lot of iteration.

You also control which collisions count. Restricting triggers to boundary contact gives a measured, deliberate performance. Allowing ball-to-ball contact in a spawner run produces something closer to a cascade, where the melody races through as the population climbs.

A MIDI file is a list of notes with pitches, durations and ordering already separated out. Feeding one in means the simulator knows exactly what note comes next and can play it cleanly at whatever moment the collision happens. This is the more reliable of the two paths and it is what to use if you have the choice.

An audio track has to be analysed into playable steps first, because a mixed recording does not arrive pre-separated into notes. That works and it is the faster option when you only have the finished song, but the result is closer to triggering slices of the recording than to performing the composition.

For a recognisable tune where each note should land clearly, MIDI is the better input. For texture, atmosphere, or a track the audience will recognise from its production rather than its melody, an audio upload is usually enough.

The failure mode of collision-driven audio is a busy stretch where twenty impacts land inside a second. Played naively that is twenty overlapping samples, which reads as a burst of noise and undoes the effect entirely.

Sound overlap prevention caps how many voices sound simultaneously, dropping the excess rather than letting them stack. Smooth multi-sound playback takes the opposite approach, keeping simultaneous notes but managing their envelopes so they stay individually audible. Which one to use depends on whether the dense section is the point of the clip or just something it passes through.