Composing for AI Voice Model Choir

The rapid growth of generative AI capability has not left singing voice synthesis and voice transformation untouched. This paper details compositional experiments in the use of AI voice conversion models outside of their normal intended purpose. Deepfake vocal models emulating well known celebrity singers have been used for singing voice substitution in popular song to make new songs with thevocal timbre of already established voices, or for new forms of mash-up. In the current paper, however, more experimental vocal improvisation, or non-vocal sounds, are used as the guide track for singing voice substitution, often multitracked across multiple voice models, leading to some interesting experimental AI choral music. The current technology for singing voice substitution is reviewed, and more radical vocal model experiments detailed. A suite of six studies, the ‘Music for Celebrity AI Voice Model Choir’ are detailed, and ethical issues discussed.