Ingest podcast feeds, crowdsource hashes of whole and partial sections of the downloaded audio, which should be a good start to auto-tag dynamically inserted ads.
For non-dynamic ads, provide an interface to manually identify their start/end, and publish for others. The same interface could be used to add chapters and other metadata.
Then you’d just point your podcast app to an RSS feed you self host.
I propose Listenarr, unless this has already been taken.