Learn · Generative music

What is generative music?

Generative music is music that a system of rules writes while it plays, instead of a recording that is played back. A composer decides the sounds, the harmony it may use and how phrases and sections behave; each performance is a new realisation of those rules.

Raumo is an iPhone app that composes ambient music this way, on the device, as you listen.

Updated

Orbit · procedural artwork from Raumo

Rules versus recordings

A recording fixes every moment of a piece. Play it again and you hear the same sound, sample for sample. A generative piece stores instructions instead: which notes belong to the key, how a melody may move, how long a section lasts, when to leave silence. Each time it plays, the system makes those decisions again, with chance working inside the limits the composer set.

The idea is older than computers. Composers wrote music that depends on dice or on the choices of the players, and in the 1970s Brian Eno layered tape loops of different lengths so that their combinations kept shifting. Eno popularised the term “generative music” in the mid-1990s, when software could make those decisions live. Bloom, the iPhone app he made with Peter Chilvers in 2008, brought it to phones.

Why a loop is audible and a composing system is not

A loop is a recording that starts again when it ends. However long it is, it has a period: the same gust of wind, the same cough in the background or the same chord change returns after the same number of seconds. People notice such recurrences quickly, especially when they listen for an hour or leave the sound on overnight. Long files and crossfades hide the seam, but not the pattern.

A composing system has no period. Its next phrase is worked out from the harmony it is in, what it has just played and a stream of seeded chance, so there is no file to come round to the beginning of. What stays the same is the character: the instruments, the key, the tempo and the kind of movement.

Not everything in Raumo is composed. Its nature layers are field recordings, released under CC0 1.0, that play as loops, and the Flow playlist uses recorded drum loops. The music above them is composed.

How Raumo writes harmony, melody, rhythm and form

Raumo’s composer is rule-based and runs on the iPhone. It uses no generative AI model and no online service. Its parts work together:

A starting point
Each of the 96 soundscapes sets a world’s instruments, a key, a tempo and a form. A seed, a starting number for its chance decisions, makes each performance its own.
Harmony
Chords follow routes written for each world, over the seven degrees of the scale and a limited vocabulary of chord colours. When the chord changes, notes move to the next chord; pitches never slide between them.
Melody
Every 20 to 40 seconds the composer plans the room’s next stretch as a whole: which voice leads, which motif it states, where the phrase rises and where it comes to rest. It then writes the melody, the answering voices and the arpeggios of that stretch together and keeps the version that scores best on one musical measure. Motifs come back with their own notes and rhythm, the way a theme returns.
Rhythm
Notes are spaced by evenly distributed patterns of onsets and rests, phrase by phrase, so the voices breathe instead of ticking.
Form
Sections thin out and fill again over minutes; density and silence are part of the plan. The notes are written a moment ahead of what you hear.
The orchestra
The notes are played by ports of Mutable Instruments Elements and Plaits, by sampled instruments and, where a soundscape has them, by field recordings, each from its own place in a room around you.

The app’s Engine view shows these modules working while the music plays. How it works follows the signal from a world to your ears.

What “never the same twice” honestly means

Generative music and AI music

The two are often confused. Generative music, in the sense used here, follows rules a person wrote: you can read them, and the same seed gives the same notes. AI music is usually made by a model trained on large amounts of existing music, which produces new audio or notes from what it learned.

Raumo is generative and not AI: no model writes its music, nothing is generated online, and its artwork is procedural, drawn by code rather than taken from photographs or image models.

Hear it

These are recordings of Raumo’s audio engine, made so the music can be heard on the web. A recording is fixed, so these are the one thing on this page that plays the same way every time. In the app each one is composed anew.