Judith Deschamps, Hugo Scurto, 2025
installation, generative AI, sound
In writing workshops organized at the Furtado Heine residential care home, elderly people trained a language model with their words, to co-write with it. All conversations were recorded, but only the spoken parts were kept. Sonic “fills”—breathing, bodily noises, silences—were not useful for the dataset. The sound artwork Silentia recovers these neglected fragments to train a neural network that generates sound textures illuminating another form of presence: nonverbal, corporeal, ordinary. Unlike algorithms that usually favor language and efficiency, this approach transforms technological rejects into a sensible archive, drawing attention to sonics that are usually excluded from generative AI.