Composing with an AI-Generated Sound Corpus: Reflections on My Computer's Interpretation of Falling Down

This paper reflects on the creation of My Computer’s Interpretation of Falling Down, a composition developed using an AI-generated sound corpus. Rather than using AI solely as a source of raw material, the work explores how generative models can also shape compositional structure and relationships between sounds. The process involved generating a corpus of motion-related sounds via text prompts submitted to a text-to-audio model, then organizing and sequencing those sounds through feature-based clustering. The result is a piece shaped through interaction—between language, system behavior, and listening. Motivated by curiosity about human–machine collaboration, the work explores how this approach might not only shape musical form but also reveal how a black-box generative model interprets a constrained topic through its underlying biases using language that is meaningful to humans, rather than adjusting abstract model parameters.