Synthetic Ornithology: Machine learning, simulations and hyper-real soundscapes

Frederick Rodrigues

Synthetic Ornithology: Machine learning, simulations and hyper-real soundscapes
Image credit: Frederick Rodrigues
  • Format: oral
  • Session: papers-5
  • Presence: in person
  • Duration: 15
  • Type: long

Abstract:

This paper presents Synthetic Ornithology, an interactive sound-based installation that uses machine learning to simulate sonic representations of localised Australian ecological futures, extending work in soundscape composition to engage in a speculative domain. Central to Synthetic Ornithology is a bespoke ML model, Environmental Audio Generation for Localised Ecologies (EAGLE), capable of generating high-quality, birdsong-focused soundscapes, up to 23 seconds in length. This paper outlines the development of the installation and how its design aims to influence audience perception of the sonic content of the work, extending established practices in NIME and sonic arts to a parafictional approach, and hyperreal aesthetics. Additionally, the paper examines the design and capabilities of the EAGLE model, and reflecting on how generative tools are positioned within a creative context, re-imagines the technical processes of training and configuring ML models as sites of artistic authorship in an expanded creative audio practice.