Musical Synthetic Media

What happens to cultural analysis when fans can synthesise the voices of their favourite musicians? WP1 examines the fast-developing field of deepfake pop as a new species of musical synthesis, tracing its practices, aesthetics, and historical antecedents across popular and art music.

The public debate around musical synthetic media has been dominated by moral alarm and naive realism: a framework of real versus fake, authentic versus inauthentic, that forecloses richer understandings of what these productions do and mean. WP1 moves beyond this framing. Through three case studies, it asks what new skills, divisions of labour, and categories of music production vocal deepfakes have precipitated; how fan-creators, named artists, and labels negotiate the boundary between cultural value and clickbait; and whether the metaphor of synthesis better describes the techne of deepfake pop than sampling.

Case studies

WP1a analyses anonymous producer Dae Lims' 2023 completion of the Beach Boys' unfinished 1967 album Smile. Already a cult classic, this synthetic-media production participates in a claim to authenticity rather than artifice, offering itself as a truer realisation of Smile's promise than the group's own efforts. WP1a situates this against the longer history of fan-created Smile mixes, from cassette tapes in the 1980s to the web in the 1990s to generative AI in the 2020s.

WP1b addresses the new aesthetics and citational practices around vocal cloning in hip hop. Where fan-producers and star musicians alike make use of vocal clones, WP1b examines both amateur and professional instances, assessing how public opinion diverges across these cases and how moral, legislative, and political-economic forces shape the genre from within.

WP1c asks how vocal cloning reshapes our understanding of the longer history of vocal synthesis. Beyond the canonical story of Homer Dudley's vocoder, WP1c traces the many subsequent vocoders oriented toward natural-sounding speech rather than robotic effects, following these techniques from mid-twentieth-century corporate R&D to twenty-first-century browser interfaces. A key case study is Charles Dodge's Any Resemblance is Purely Coincidental (1980), an understudied work that faked the vocal performance of opera singer Enrico Caruso using linear predictive coding.

Methods

Archival research, interviews and oral history, and practice-research. The PI and PD1 will reconstruct the 1976 implementation of the linear predictive coding used in Any Resemblance, both to enable closer analysis of the piece and to illuminate vocal cloning's longer history.