Remix.run Logo
▲ makefu a day ago

Anyone is aware of software which does the same but for podcasts? I've had built a downloader by qwen38 (https://cgit.euer.krebsco.de/makefu/podfetch) which does so via multiple programmatic approaches (keywords, sponsorblock-ml, neuralblock) but actually i think crowdsourcing the solution is the best approach here.

▲rdwallis a day ago | parent | next [-]

Downloading the podcast via a computer or vpn in a low income country will remove almost all dynamic adverts (which is most podcast advertising)

I lived in Tanzania for a few years and most podcasts wouldn't play adverts while I was there.

▲mgrandl a day ago | parent | prev | next [-]

Crowdsourcing doesn’t work for Podcasts because most providers cut in ad segments based on your location. Those differ in length so the sponsorblock approach would not work. I slopped up a small go app that uses whisper API to transcribe and then uses Jev-compatible API to classify ad segments. Works great and costs like $0.01 per episode.

▲xnx a day ago | parent | next [-]

I've wondered if it would be possible to crowdsource tiny -but distinct- segments of audio that bookend each ad break so clients could remove any ads inserted between those points.

▲eik3_de a day ago | parent | prev [-]

Would you mind sharing a little bit more about that jev approach? I'm building something similar and would be interested in trying it out. (Don't worry about pasting some labelled AI generated description)

▲mgrandl a day ago | parent [-]

[flagged]

▲xnx a day ago | parent | prev | next [-]

Like the sibling comment, I rolled my own (local Parakeet + Gemini), but this uses the same approach and is actively developed: https://github.com/podly-pure-podcasts/podly_pure_podcasts

▲makefu a day ago | parent | next [-]

I like the approach it acts as a kind of RSS proxy which cleans up the original feed. Also just offloading the task to the LLM API is really something. Thanks for sharing and i hope your SaaS works out for you!

▲notrealyme123 a day ago | parent | prev [-]

Would love to look at it, but GitHub gives 504.

▲xnx a day ago | parent [-]

I got the same, but the page is back up again now.

▲briHass 14 hours ago | parent | prev [-]

Ironically, the best answer is to see if the podcast is available on YT and use sponsorblock.

I went as far as to write a YouTube downloading app, that uses ytdlp, but presents the downloaded mp3s in a feed that can be picked up by antennapod.