Live Captioning FAQ | Hacker News Reader

Live Captioning FAQ

whitecoatcaptioning.com

29 pointsmooreds14 comments

I've been to a few events where White Coat have been doing the captioning, and they're very good. I often find myself reading the live captions even whilst I'm listening to the speaker, I think my brain sometimes finds it easier to follow along with (e.g. to lighten the cognitive load a little if the speaker has a strong accent which I'm not tuned-in to very well). Live captioning is highly recommended for all sizeable events, and White Coat seem to be pretty good at it (I have no relationship with them, just encountered them a couple of times).

I saw them at RemixConf and was completely blown away by how well they nailed it. Technical terms, acronyms, and even non-technical quirks of the speaker (like one who said “regular-degular HTML”) and the captioner didn’t miss a beat.

(Similarly no relation to them, but I passed their contact info along to our events team after seeing them in action and hope my company followed up with them.)

Live captioning (not subtitles) is quite difficult. We'll be contacting these guys next year, as FCC-level captioning is a requirement for some of our customers...and as time goes on, the FCC will probably require captioning for VoD and live streams.

I'm surprised nobody's hit YT etc for violating the ADA...although they do have automatic subtitling, which may satisfy the requirements.

>I'm surprised nobody's hit YT etc for violating the ADA.

Does YT fall under ADA? It's not a broadcasting platform. It is user generated content. Not sure if that falls under ADA either.

I've been part of teams scrambling to populate back catalog content with captioning (not-subs. they're not the same) for existing streaming content. Lots of 3rd parties involved, lots of buying of software to handle volume when 3rd parties become too expensive, etc. The cheat was to turn subtitles into captioning, but not doing the actual information for the hard of hearing.

Youtube asks you some questions when uploading, and the detail they have clearly indicates that they are trying to avoid closed-captioning requirements (and their best-effort auto-captioning is trying to be good enough that they don't get hit with requirements).

I would NOT be surprised if they go back and manually correct captions on very popular videos.

https://support.google.com/youtube/answer/3068031?hl=en might be a place to start

By reading the homepage, I can gather that this is a service that offer onsite or online captioning by HUMAN, not by machines. Being a live interpreter at medical conferences myself, I would say these people have mad skills. Listening, translating, and typings live? That's just pure talent.

The “translation” is conversion of the shorthand they type into regular English words and phrases. They’re not listening to a different human language and translating it into English.

Doing any live captioning well is challenging, doing it when there is a lot of jargon and variety of speakers even more so.

My bad. I was confusing between captioning and translating to another language

I believe there are companies that will offer that, as I've seen "live" English captioning of someone speaking another language.

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.

The 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.

If 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.

I recently saw this great plugin for OBS which uses Google Speech Recognition api to stream captions to twitch along with your video. Not sure about the quality or delay.

https://github.com/ratwithacompiler/OBS-captions-plugin

https://obsproject.com/forum/resources/closed-captioning-via...

Of all the transcribers we used, google STT was the worst one.

GIGO.

That said, their translation was pretty good, but the best is apparently DeepL.

Reply on news.ycombinator.com