Marco ArmentRE: https://podcastindex.social/@francosolerio/116285636972773274
I love getting submastodoned
and I'll have you know that they all have that delightful new-Apple-computer smell, thank you very much
Jarrod Blundy@marcoarment With the performance that @matt_birchler‘s getting out of Quick Subtitles, I was a little surprised that you didn’t go the local transcript route. Seems like it would account for the DAI problem too. But you like centrally supplying the transcripts better?
bubbajet@jarrod @marcoarment @matt_birchler This is actually discussed in the ATP podcast. As a nerd who is in way out of his depth, I found the discussion fascinating. Recommend.
ATP #683
@bubbajet @marcoarment I’ll give it another listen! Maybe it went over my head the first time. But if over apps are able to do it (dunno how well…), maybe it’s more viable than first it seemed?
Matt Birchler@jarrod @bubbajet @marcoarment FWIW Overcast does have local transcription built in, but it's slower and way more processor intensive than having a transcript ready to go right away.
Takes about 2 minutes of maxing out my iPhone 17 Pro to generate one for an hour+ episode, and even longer on older phones.
Also, Overcast supports iOS 18, which can not use this local transcription, so doing it sever-side gives those users the feature as well, which is nice.
Connor Graham@matt_birchler @jarrod @bubbajet @marcoarment Personally, I also appreciate Marco's approach because, even though these devices are individually power efficient, having thousands of devices separately transcribe a new episode of a popular show is just orders of magnitude more total energy use than transcribing it once and sharing it.
@connor_g @matt_birchler @jarrod @bubbajet And given how long it takes, it can't be reliably done in the background, so it's a poor user experience to NEED to transcribe every episode manually before you can use transcript features.
It also restricts features like search.
@marcoarment @connor_g @matt_birchler @bubbajet That all makes a lot of sense! Thanks for enlightening me. 😁 I’m curious how Castamatic’s approach will pan out for its users.