Podcasts often dynamically generate ads at the point of download, making the SponsorBlock-approach unviable: since the media is expected to be variable-length you can’t store media positions that map to advertisement segments.
You could potentially match on audio, though – look for the 15 seconds of podcast audio preceding the ad, and the 15 seconds following it, if folks reported it in a sponsorblock way.
Alternatively, we could build a shazam-style database of 30 second podcast ads, then skip them when they’re identified. There isn’t much variety out there.
Ah interesting, that just shows that I have successfully avoided subscribing to those. The few ones with ads I listen to get the ads read by the hosts like in YouTube videos.
like others have pointed out, this makes automatic detection easier not harder if true. Just cut the segments that move around, audio analysis and even transcripts (they don’t have to be good! just good enough to identify missing segment!) are pretty mature.
Podcasts often dynamically generate ads at the point of download, making the SponsorBlock-approach unviable: since the media is expected to be variable-length you can’t store media positions that map to advertisement segments.
You could potentially match on audio, though – look for the 15 seconds of podcast audio preceding the ad, and the 15 seconds following it, if folks reported it in a sponsorblock way.
Alternatively, we could build a shazam-style database of 30 second podcast ads, then skip them when they’re identified. There isn’t much variety out there.
Ah interesting, that just shows that I have successfully avoided subscribing to those. The few ones with ads I listen to get the ads read by the hosts like in YouTube videos.
like others have pointed out, this makes automatic detection easier not harder if true. Just cut the segments that move around, audio analysis and even transcripts (they don’t have to be good! just good enough to identify missing segment!) are pretty mature.