Sifting sound interactive extraction, exploration, and expressive recombination of large and heterogeneous audio collections

Abstract

How can the tasks of searching, finding, and combining sounds be transformed into acts of exploration, discovery, and design? In the age of big data, recorded sound abounds in both volume and variety. When data is used not for the passive extraction of knowledge, but as aesthetic material for creative media production, the human role in curating and shaping its contents is even more essential. While access to this growing expanse of material is nominally more available than ever, this access is encumbered by the time, technical facility, and experienced aesthetic judgment it takes to effectively scan large amounts of unedited and unstructured media for contextually interesting elements, discover connections among them, and recombine them intentionally into new compositions, leaving extraordinary creative resources unexplored. This thesis offers a new platform for audio exploration, discovery, and creativity that addresses this problem by developing tools that build on, rather than replace, human creative judgment in extracting interesting samples from long audio recordings, navigating across a large number of these to form subcollections, and guiding sounds back into expressive musical forms.