Musical AI Voices: Facts, Concerns and Experimental Musical Practices with AI Voice Tools

Licentiatavhandling, 2024

As early adopters of technologies, artists are uniquely positioned to probe and explore the potentials and problematics that these tools afford and pose to their artistic practice. This is especially applicable to AI voice tools, which present unique challenges to the "value" of human voice and voice data; to our understandings of human vocality in the age of deep learning innovation; and how we work with AI voices in an experimental musical practice. This thesis investigates the intersections of AI voice tools and experimental musical practices, examining how artists critically engage with--and are implicated by--AI voice tools that clone, parse and synthesise human voice and speech. 

Within this topic, we address the technological facts and the societal implications and concerns of generative AI voice tools--encompassing deep learning voice models and speech toolkits--which offer unique artistic potentials to work with a feasibly unending palette of generated vocal sounds. The fidelity of these tools are continually advancing, and are increasingly being utilised within artistic practices. 

Our motivations in this thesis are grounded in an exploration of AI voice tools' potentials and problematics, which span a range of both pragmatic technology facts and societal concerns, and navigate interdisciplinary boundaries. The nature and pace of deep learning developments is such that we presently lack methods of visibilising and critiquing the potentials and problematics of AI voice tools used in musical contexts. Further, we are in a unfolding period of investigating the wider socio-technical implications that are constructed through these potentials and problematics within musical practice. This thesis therefore explores the following research questions:

What methodologies assist in visibilising the multifaceted potentials and/or problematics of AI voice tools used in musical contexts?; What wider socio-technical implications occur through these potentials and problematics within musical practice?; and What shifts occur within an experimental musical practice when critically exploring AI voice and speech tools?

Seeking to answer these questions, this thesis develops methodologies for--and chronicles--interdisciplinary practical and theoretical engagements with AI voice and speech models. Further, it discusses and formulates practical methods for feminist and interventionist analysis, and the development of--and performance with--AI voice and speech tools in experimental musical settings. Questions on how to visibilise the potentials and problematics of AI voice tools in musical contexts are foregrounded, alongside explorations into the shifts that occur within experimental musical practices when engaging with such tools. 

This thesis contributes with: 1) a novel analytical method for the critical analysis of artworks featuring musical AI voice tools; 2) the establishment of interdisciplinary perspectives as integral to understanding the use, cultures-of-use and implications of voice and speech AI tools in musical applications; 3) a Research-through-Design account of developing and performing with a series of AI voice models in a live music performance; and 4) a research stance on experimental musical practices as enabling the formation of new understandings of human and AI-mediated human vocality.