AI-generated voices are now virtually indistinguishable from real human voices, according to researchers from Queen Mary University of London.