STRATA II

STRATA II is a work of sonic archaeology. It is the second piece of Strata, an album in which I research the sonic archaeologies.

I dedicate it to my friend Halil İbrahim Tutar, with whom I walked this city so much, and through whom I came to know its atmosphere.

What interests me here is a shift in what a composer does. The work is less and less about inventing material, and more about surveying an archive and refining what is already in it. Whatever survives of a city’s sound reaches us damaged: it has passed through recording media, through decisions about what was worth keeping, through generations of reproduction that flattened it into something standard. I wanted to treat that damage as the subject rather than as a defect — to follow sound as a historical remnant and ask what journey it has taken through time.

The piece is twelve channels and ten minutes, built from field recordings of Istanbul. It does not stay fixed. While it plays, three neural models process it in real time, each feeding its own output back into itself, so the erosion accumulates as you listen. None of these models has ever heard this city: one was trained on magnetic fields, one on a hand drum, one on ten thousand stock sound-library loops. They can only rebuild Istanbul in terms they already possess, and whatever they cannot hold falls away.

That is the excavation. What is left after each pass is not the city, but what a machine could still carry of it — and the distance between the two is what the piece is about.

The system listens to its own output and decides how far to take the erosion. There is no performer, and it comes out differently every time it runs. What you are hearing is one run.

Technical note

The engine is three instances of RAVE (Realtime Audio Variational autoEncoder, Caillon & Esling, IRCAM ACIDS), running in Max 9 through the nn~ external, each wired as a feedback loop so that degradation accumulates continuously rather than in fixed steps. A listener module measures the spectral drift of the system’s own output against the untouched original and corrects the depth of erosion, so the piece can neither collapse into noise nor freeze. The three voices are encoded into a third-order ambisonic field and decoded to twelve loudspeakers; their elevation is tied to the depth of erosion, so the copy rises above the original as it takes over.

No model was trained on the source material. This is deliberate.

An earlier piece, Strata, excavated the same city by hand.

For the twelve-channel version, the Max project, or the technical documentation, please get in touch.