A listening post that mishears the lagoon, on purpose
A hydrophone in the lagoon, a transcription model well outside its training, and the poetry of a confident wrong answer.
A hydrophone dropped off a jetty, feeding a speech transcription model that has never heard water. It produces a continuous stream of confident, entirely wrong sentences.
The piece is about what a model does at the edge of its competence, which is not to fall silent but to keep answering. Printed on a receipt roll, it reads like someone talking in their sleep.
The technical work was mostly conditioning: a hydrophone into a preamp into a filter that keeps the band the model expects and throws away the rest. Without that it transcribes nothing at all.
It printed roughly forty metres of paper over six weeks.
Built on 03 · Live VJing and live visuals with TouchDesigner. ₦85,000. See the course
