I wonder if you could train some machine learning model using the data from SponsorBlock and achieve good results on podcasts as well. That way you wouldn't be

I wonder if you could train some machine learning model using the data from SponsorBlock and achieve good results on podcasts as well. That way you wouldn't be dependent on a crowdsourced online database for your offline listening. Alternatively, even creating a transscript using something like [1] and scanning for words like "sponsor", "ad" or specific company names might already be a good enough heuristic. [1] https://github.com/alphacep/vosk-api https://github.com/alphacep/vosk-api