It’s called dynamic ad insertion. Online radio stations use it all the time for localized ads based on IP address. Did some work for a set of radio stations in the way of building a diagnostic web player for the metadata in HLS streams debugging their new ad system. (Using HLS.js which I have to say I’m grateful for as it’s a brilliant library)

So they don’t have to stitch together one new audio file per local but place metadata markers in the audio file metadata that prompts ad insertion. These usually have a marker for length and the systems handle number of plays and placement spots. Don’t know as much about that end. HLS being a chunked format would allow the files to be stitched back together on the client end as well with ads in place.

Wouldn’t be surprised if they’re doing this. Also wouldn’t be surprised if many Podcasts are broadcast via HLS as it’s native in Apple software and Apple devices. Someone else with more knowledge would have to confirm or deny that, though.