Deep fakes are audio or video outputs of generative adversarial networks (GANs) that are trained on real audio/video and can be difficult to distinguish from that produced by people. Famously this can be used for negative purposes, such as appearing to be pornography produced by public figures, but it can also have positive uses, such as the creation of artificial voices for patients who are unable to talk without assistance, such as people who have lost their voice following surgery or illness, or for voice artists to increase their output.
One use of this technology would be a generalised voice simulator, trained on the result of a crawl of the internet for voice recordings, appropriately clustered. Let us suppose that because of his distinctiveness and frequency of appearance on the internet one such voice is trained on lots of Brian Blessed and sounds quite like him, but there was no human input to ensure this happened. Would Brian Blessed have any claim against an entity distributing such a voice simulator, based on the output being a derivative work of his voice recordings among others?
