So far, I still wasn’t enamored with the auto-generated subtitles for that, because it’ll occasionally choose the wrong word, which is then worse for me than just hearing the unclear speech myself.
But yeah, we’ll have to see how technology advances. I assume, LLMs can guess the correct word based on sentence structure, so there’s probably still a good bit of room for improvement.
Colleagues have also been working on some speech-controlled UI and they do report huge differences in how different models can deal with bad audio quality, so that seems like things are going forward, too.
Yeah, I’ve seen that from YouTube’s auto-generated subtitles. If it’s having a really hard time understanding someone I also prefer having it off, but I’m okay with a few mistakes here and there.
So far, I still wasn’t enamored with the auto-generated subtitles for that, because it’ll occasionally choose the wrong word, which is then worse for me than just hearing the unclear speech myself.
But yeah, we’ll have to see how technology advances. I assume, LLMs can guess the correct word based on sentence structure, so there’s probably still a good bit of room for improvement.
Colleagues have also been working on some speech-controlled UI and they do report huge differences in how different models can deal with bad audio quality, so that seems like things are going forward, too.
Yeah, I’ve seen that from YouTube’s auto-generated subtitles. If it’s having a really hard time understanding someone I also prefer having it off, but I’m okay with a few mistakes here and there.
It’s great that YouTube offers subtitles and I believe they’re better than no subtitles in most cases, but man do they suck in many cases.