Forever Frequencies is an installation that computes a personal melody from a visitor’s memory, using their own past and imagined future as the raw material for a one-of-a-kind musical composition.
The installation lives through the materiality of the timeless music box, evoking early childhood as a moment of both wonder and simplicity, where empty spaces begin to fill with music and memories. Visitors approach the music box and answer a simple questionnaire about a music-related memory: where did it take place, when, and with whom? The narrative then shifts to the future, prompting them to imagine a future memory, a liminal space in time where the past reflects itself in what is yet to come. Taking in these answers, the installation computes a personal, one-of-a-kind melody, their own “forever frequency”, using a local-based AI model. With each visitor creating their own tune, the installation builds into a kind of joint symphony of sound and reminiscence.
In addition to the main piece, we collaborated with artists and thinkers from different fields: architect Kengo Kuma, artist and composer Maria Arnal, musician and sound designer Suzanne Ciani, music producer Max Cooper, electronic musician Richie Hawtin, media artist Daito Manabe, muralist Elian Chali, chef Ada Parellada, and conductor Kevin Rhodes, each contributing a music-related memory that generated their own forever frequency, on display for visitors to hear and read.
The installation was on display from May to August 2025 at the Barbican Centre in London, as part of “Feel the Sound: an exhibition experience on a different frequency”, and will continue to travel – next to the Museum of Narrative in Japan – each time relying on physical, on-site local AI systems.
The core of the piece uses a Murobox – a MIDI-controlled mechanical music box – to play melodies shaped by each visitor’s responses. The composition pathway, fully crafted by our team, moves through several steps of decoding and translation: the user’s answers influence musical characteristics (tempo, range, key) which serve as parameters for an LLM (Large Language Model) to generate a melody in ABC musical notation, later converted into a MIDI file. A second LLM model then analyses the generated song to output a text contextualising the musical memory and describing its emotional significance to the visitor.
Rather than relying on the inherent randomness of generative AI, we designed a system to maintain a higher degree of control over the musical outcome, going beyond current conventions to ensure each composition carries genuine intentionality. The LLM server the artwork runs on is local, only available on-site at each installation, which was a deliberate choice we made to minimise ecological footprint and avoid dependency on remote computation. The composition pathway itself was fully designed by our team, ensuring the system translates human experience into musical form with artistry and structure rather than chance. We remain critical and transparent in our decisions about the tools we use and their impact, testing a variety of technologies before arriving at the final form, always guided by meaningful artistic needs. Every iteration of the installation relies on on-site hardware, reclaiming the analog and purely human aspect of shared memory rather than offloading to cloud infrastructure.
Project Page | Domestic Data Streamers | Instagram
