the frequency-domain characteristics of the human voice are difficult for humans to fake because they arise from the shape of the speaker's individual vocal tract, but machines could always fake them even without statistical methods (see vocoders and autotune); the time-domain characteristics of the human voice like prosody and intonation are easy for humans to fake, and statistical methods give computers a superhuman level of recall and diligence in applying them.
the fact that these were ever deployed is an indictment of the tech industry.
