Just when dubs were getting to a point of not being too bad. Subtitles, especially CC have felt especially bad recently. I would rather wait an extra week or two for good sub or dub then have it done poorly or flat out butcherd with A.I…
Honestly that’s because speeding up localizations by having the first pass be machine-made is not something that waited for GenAI to happen. It’s been going on for a while using good old machine translators.
Now, Google Translate and similar tools have been reliant on machine learning for ages, people just weren’t freaking out about it because “AI” hadn’t gone viral. It’s been weird to watch this sort of thing play out.
FWIW, if they are using the same loc workflow and genAI works better than good old machine translations for a first pass go ahead and do GenAI. From what I’ve seen casually it’s not necessarily faster or more reliable, but I’m not working on loc professionally. Maybe that’s what he means when he talks about using it in “backend processes”?
Just when dubs were getting to a point of not being too bad. Subtitles, especially CC have felt especially bad recently. I would rather wait an extra week or two for good sub or dub then have it done poorly or flat out butcherd with A.I…
I’ve noticed it on non-anime as well. It isn’t even good at doing English CC when the spoken language is English.
Honestly that’s because speeding up localizations by having the first pass be machine-made is not something that waited for GenAI to happen. It’s been going on for a while using good old machine translators.
Now, Google Translate and similar tools have been reliant on machine learning for ages, people just weren’t freaking out about it because “AI” hadn’t gone viral. It’s been weird to watch this sort of thing play out.
FWIW, if they are using the same loc workflow and genAI works better than good old machine translations for a first pass go ahead and do GenAI. From what I’ve seen casually it’s not necessarily faster or more reliable, but I’m not working on loc professionally. Maybe that’s what he means when he talks about using it in “backend processes”?