Her earlier videos have her normal voice. It wouldn't be that difficult to bypass the TOS of CursorAI to pull the audio out from all of her videos, hash them, analyze the pitch of them all, then take the ones that have her normal voice, do an analysis on it compared to the rest, and figure out what the filter is, and reverse it, and apply it to whatever video you want.
Only real issue is the space involved. As long as you rip the audio from the videos FIRST so CursorAI never knows the content of the videos, and you process EVERYTHING locally you might be fine.
In fact, it would be safest to take her original video ASMR or cooking videos, compare it to segments of her new videos with the naughty bits cut out, find the filter, reverse it, then have a script you can apply on all the videos.
Weekend project if anyone is interested.