Abstract: The increasing ability of deep learning models to produce realistic-sounding synthetic speech poses serious problems for privacy, public trust, and digital security. To counter this danger, ...