4 ms·
I oftentimes wonder if there’s any way to have a system where a ML inferencer does the majority of the captioning work and a volunteer can pickup the various sp
by 1-6 4y ago
I oftentimes wonder if there’s any way to have a system where a ML inferencer does the majority of the captioning work and a volunteer can pickup the various spots where it went wrong.
Basically, each word that has a low prediction value can show multiple choice options and the volunteer simply chooses the correct word on a number pad.
Hiring a real person sounds expensive so I’m trying to target a 100-500 people audience with a service that costs $0-10 per day.
- bombcar 4y agoThe problem is the delay - if you're going live, the human barely has time to notice the error before it's off the screen. If everything is delayed a bit, it could work. What I've seen is ML for the "live show" and then you fix the transcript for the recording.
- manv1 4y agoIf you pay attention, you'll see that the NFL captions are delayed by about 4-5 seconds, sometimes longer. The FCC doesn't mandate any realtime window for live events. And when lots of people are talking you'll run out of screen. I wonder if a ticker would be better than the traditional caption overlay.