Stairwell Light

Score / Sound

Stairwell Light is a collection of short piano works created through a multi-stage generative process.

Each piece begins with the same simple request:

“Compose a short work for solo piano …”

What follows is only a small variation each time.

Sometimes the language model is given as much freedom as possible:

“Make all other musical decisions yourself.”

Sometimes a single formal or psychological impulse is added:

“Let later material retain a perceptible memory of earlier material.”

“Do not decide on the form in advance; let it emerge from the material.”

Other instructions explicitly address an imagined human listener:

“Favor musical ideas that are likely to feel expressive and rewarding to human listeners.”

or:

“Compose with the expectation that a human listener should want to hear the piece again.”

These sentences specify neither key nor harmony, form nor style. They simply establish a framework. The actual musical decisions are left to the model.

Score

The first form of each work is a score.

A large language model composes the piece symbolically and writes it entirely in LilyPond. It decides on pitches, rhythm, voices, registers, harmony, texture, dynamics and formal development.

The score is therefore not supplementary material for the recording. It is a musical artefact in its own right and the compositional origin of the piece.

The model has extensive knowledge of musical structures and conventions. Yet it has no human history of listening. It has never played an instrument, experienced a concert hall or associated personal memories with music.

This very difference is part of the project:

What happens when a system can generate music structurally without experiencing it in a human way?

And how do even very small changes in language alter the space of musical possibilities that emerges?

From Score to Sound

The second published form of the work is not simply a performance of the score.

Since the generative audio model cannot read a LilyPond score, the composition must first be sonified by technical means. The score becomes MIDI, which is then rendered as a simple piano audio file.

This intermediate stage has no independent artistic status within the project. It is a necessary interface between two different models: one working symbolically, the other acoustically.

The audio is then passed to a generative sound model.

What emerges is more than a sonic refinement of the MIDI rendering.

At times, the model intervenes deeply in the music. Arpeggios may expand to include additional notes, textures change, tempi shift, proportions alter, and gestures are extended, condensed or given new weight.

The sound remains clearly connected to the score, but it is not identical to it.

A second generative act thus takes place.

CODEX makes symbolic decisions.

SUNO makes acoustic decisions — and, at times, further musical ones.

The boundary between composition, interpretation and transformation is deliberately blurred.

Score / Sound

The subtitle Score / Sound names precisely this dual structure.

The Score reveals the symbolic composition.

The Sound reveals what happens when that composition becomes the context for a second generative model.

Both forms belong to the work, but they are not the same.

Motifs persist, harmonic relationships leave traces, and formal developments remain recognisable. At the same time, new notes, different temporal emphases and altered musical gestures emerge.

The work therefore exists not only on both sides of the slash.

It also exists in the difference between them.

System Authorship

My own role is not that of a traditional composer.

I do not decide on every note, nor do I perform the pieces myself.

I design the system.

I formulate the starting conditions, determine the degrees of freedom, organise the transitions between models, listen to the results, make selections and decide which versions are published.

This role can be described as system authorship.

The models generate possibilities.

The human author establishes the process, listens, compares and decides which of these possibilities become part of the work.

In doing so, I am both the producer and the system’s first listener.

This act of listening is itself an important part of the process. The actual music is not fully planned in advance. It can surprise, unsettle or convince.

Selection thus becomes an artistic act in its own right.

Stairwell Light

The album takes its title from the piece with the most open instruction:

“Make all other musical decisions yourself.”

A stairwell is a transitional space. It is neither a starting point nor a destination. We pass through it from one level to the next.

This music, too, moves through transitions:

from language to notation,

from notation to technical sonification,

from sonification to generative sound,

from machine-generated structure to human listening.

The two artistically significant states of this process are published:

Score / Sound.

In the end, all of this can be analysed.

We can compare score and sound, reflect on machine authorship or examine which decisions arose at which stage.

But we do not have to.

At some point, the process recedes.

And what remains is music.