Post by @mawhrin@circumstances.run

2026-09-20 10:50 UTC

@JigmeDatse@social.openpsychology.net yes and no; it's better than nothing, but such a transcript has to be proofed by a human to be actually useful. the thread here describes some of the failure modes. (@jzb@hachyderm.io also had a bunch of very good general points in another reply somewhere in this thread, and they summarise my position rather well too).

Replies (1)

@JigmeDatse@social.openpsychology.net 2026-09-20 10:54

@mawhrin@circumstances.run @jzb@hachyderm.io Yep, I'll say that it can be terrible. I'll take a better look. What I'm seeing on PeerTube, is way better than YouTube. And I'm running an old instance. I get it may well also be "way more expensive" to do it that way, but the tests I've done, have had about 2-3 minor errors per 10 minute video. It also might be a lot easier to tweak to get better results (dangerous, as it could lead to not testing). Just your quick response is more or less where I'm at. Still not terribly happy with it heading that way, but in cases where it would either be not captioned, or a tweaked automated captioning, that's good. Manual captioning should be done, where there's resources for it.

Open ##5390254