Voice increasingly mediates artificially intelligent (AI)-enabled communication, with the expanding proliferation of conversational AI systems like Amazon’s Echo, voiced by ‘Alexa’. This research is concerned with the sound and sounding of voices (human and synthesised) in conversational AI systems and identifies that vocal profiling underpins current understandings of vocal sounding by AI and the AI industry. Vocal profiling relies on normative assumptions that risk misrepresenting individuals, negatively impacting those already marginalised. By fundamentally reorienting the discussion of vocal profiling from ‘listening’ towards ‘sounding’, the voice is given increased agency. The practice-led research created vocal imaginaries that reveal and resist vocal profiling, questioning and disrupting the dominant understanding of voice(s) promoted by AI. Situated at the intersection of sound, design and technology, this research incorporates contemporary societal discourse on identity politics, personhood, being and ecology.

Through this PhD research, a sonic speculative design methodology emerged as an original contribution to knowledge, where working with and through sound in the design process created opportunities to establish novel concepts with a turn towards co-creation. Termed ‘Speculative Voicing’, sonic thinking was successfully applied to speculative design to confront contemporary critique of the field, generating a new materialist, intersectional approach. It was developed over six participatory workshops with young people and then applied to an investigation of vocal profiling in conversational AI systems through two case study projects. Evaluation of the works took place in a further workshop with industry professionals. The evaluation found that the Speculative Voicing methodology accomplished the creation of vocal imaginaries that resist vocal profiling while creating space to apply narratives that can also reveal them.

Starting from my position as a practising speculative designer and vocal performer, I propose the Speculative Voicing methodology to work with the voice in conversational AI as a design material and treat it as an experimental singer would. Vocal potential is thus generated, building dynamic relations and entanglements to explore concepts of being and identity, supported and evidenced by the motivations of a lineage of female experimental vocalists. Guiding principles of the methodology explorations posit that the voice is always polyphonic through its materiality – co-created with many bodies, environments, matter and the AI system. The recurring metaphor of a vacuum highlights the sonic materiality of voice. Critique is constructed through reflexive analysis of the practice work that demonstrates vocal potentiality in comparison to current profiled presentations and representations of voices in contemporary conversational AI systems.

Speculative Voicing provides a practice-focused, voice-led methodology of working with the polyphonic sonic materiality of voices to reveal and resist voice profiling practices. It is intended for those who want to explore speculative design practice from an intersectional position, especially when working, or intending to work, with voice. The methodology is supplemented by accessible outputs for further impact, dissemination, and advocacy against current practices of vocal profiling. These comprise a Speculative Voicing Framework, a prototyped participatory workshop and two interactive tools to reflect and offer feedback on the understanding, design and implementation of voices in conversational AI.