The Paris Hilton voice clone video that made me question my own ears

I spent 3 years telling people audio deep fakes were easy to catch. Then last Tuesday I watched this 45 second clip where Paris Hilton was supposedly reading a slam poem about cryptocurrency. It sounded so natural to me, the little breathy pauses and the way she drags out certain words. I was ready to call it real until someone in the comments pointed out that a real audiobook of hers from 2018 was layered underneath the fake track. I pulled up both files in Audacity and the timing matched to the millisecond. That detail changed my whole approach. Now I check for source audio matching before I even listen to the words. Has anyone else been fooled by a voice clone that was built on existing recordings instead of synthetic from scratch?

1 comments

1 Comment

evan_davis3h ago

man the layered source thing is exactly what got me too. i was convinced this podcast clip of a local politician was legit until i matched the vocal fry pattern to an old interview he did five years back. it was like the ai just took his cadence and stitched new words into the same breaths. the scariest part is when they keep the original pauses and lip smacks, your brain just locks onto that as proof of humanity. now i do the same thing you do, i hunt for the source track before i even judge the content. it feels paranoid but honestly it's the only way to not get played.