“Call me” controlling - What it takes for a singer to feel comfortable using voice cloning creatively

In 2022, I recorded hours of singing to be able to synthesise vocals in my timbre using transformer models. The results are uncanny snippets of word-less vocals that match my vocal style. These output stems were used in several released songs. One of them ,“Call Me”, exemplifies an idealised workflow and system for singers to work collaboratively with new technology without fear. By looking at the requirements of this supposed ideal way of working creatively with AI, I highlight that these conditions can rarely be filled and make recommendations to preserve artist's comfort and control in work featuring generative AI.
Beyond covering the technical aspect of training a model’s on one’s voice and using them in popular music production, this paper focuses on how the perception of control can change our relationship to new technologies and their applications to one’s defining features, like a singer’s voice. This idealistic yet real use-case allows to reflect on the ethics and social ramifications of generative AI in the context of music creation.