Not I: The Voice, Identity and the Epistemic Mirage of Machine Learning

Not I is an essay film and multichannel installation that explores the problematics of contemporary vocal profiling technology. It challenges the impulse to apply statistical learning techniques to polymorphous features of human expression for the purpose of speculative reconstruction exhibited in data science. Building upon the work of scholars across sound studies and vocal studies, Not I troubles the assumptions behind attempts to distil a one-to-one mapping of voice and identity. The film centres on an investigation into, and attack on, Speech2Face, a machine learning model which attempts to generate an image of the face of a speaker based solely on a recording of their voice. We leverage the affective qualities of moving image work to present this investigation as a form of experiential critique, forcing the viewer into an affective scenario that unsettles their existing heuristics used to infer speaker identity from vocal perception.