• Get involved.
    We want your input!
    Apply for Membership and join the conversations about everything related to broadcasting.

    After we receive your registration, a moderator will review it. After your registration is approved, you will be permitted to post.
    If you use a disposable or false email address, your registration will be rejected.

    After your membership is approved, please take a minute to tell us a little bit about yourself.
    https://www.radiodiscussions.com/forums/introduce-yourself.1088/

    Thanks in advance and have fun!
    RadioDiscussions Administrators

Subtitles in TV but may affect AI in radio!

davideduardo

Moderator/Administrator
Staff member
Last night, I vaught the last few seconds of ABC’s 20/20 where the audio was “… and she plans to appeal the verdict” but the subtitle said “… and she plans to face peel her dick.”

I am assuming that subtitles on "live" shows are now done by voice recognition software. ABC’s is obviously not even in Beta testing stage yet!

This is just the beginning of what may be an AI nightmare, since AI does not really think and it can come up with word combinations that are unfortunate.

I spent a few minutes with a friend thing of word combinations that AI might not detect as wrong... such as talking about a fan doing the job of blowing air through a server farm or personal computer.
 
Last night, I vaught the last few seconds of ABC’s 20/20 where the audio was “… and she plans to appeal the verdict” but the subtitle said “… and she plans to face peel her dick.”
Could be, but where I work I've seen Closed Caption Transcriptionist's make dumb spelling errors, or mis-hearing what's being said, typing completely different words. Some of the CC folks are even overseas with a language barrier. Over the past year or so, I'd rather roll the dice on a voice recognition captioning system than some of the dumb-asses doing manual captioning, when they remember to log in.
I am assuming that subtitles on "live" shows are now done by voice recognition software. ABC’s is obviously not even in Beta testing stage yet!
Voice recognition captioning is a pretty mature technology. If you've ever turned on closed captioning on YouTube videos, you'll see examples where it's really good, even with amateur-grade mic techniques and recordings.

This is just the beginning of what may be an AI nightmare, since AI does not really think and it can come up with word combinations that are unfortunate.
Voice recognition isn't really AI-much. There are some rudimentary word assumptions thrown in, but it's pretty much a one-for-one word recognition.
 
Voice recognition captioning is a pretty mature technology. If you've ever turned on closed captioning on YouTube videos, you'll see examples where it's really good, even with amateur-grade mic techniques and recordings.
Because English has become my second language and because some regional accents are very hard for me to fully follow (New England, Deep South, Texan and Urban), I always have subtitles on.

I see strange transcripts all the time; frequently difficult sentences are entirely skipped.
Voice recognition isn't really AI-much. There are some rudimentary word assumptions thrown in, but it's pretty much a one-for-one word recognition.
My experience is far less satisfactory. While previously filmed or recorded shows such as scripted episodes and even entertainment ones like "Dancing with Ogres" or whatever they are called are fairly well transcribed, anything live or "nearly live" (recorded an hour or so before broadcast like the late night shows) are somewhere between "rough" and "terrible".
 
Lots of the closed captioning used to be done at WGBH/PBS in Boston in the bad old days.

Now it might be voice to text, but even my Google based V2T will blank out objectionable words, much like the software package this website uses.
 
Last night, I vaught the last few seconds of ABC’s 20/20 where the audio was “… and she plans to appeal the verdict” but the subtitle said “… and she plans to face peel her dick.”

It's "20/20". Are we SURE it wasn't "face peel her...."

Just joking.

I am assuming that subtitles on "live" shows are now done by voice recognition software. ABC’s is obviously not even in Beta testing stage yet!

I have ranted for 25 years to anyone who'll listen about how awful closed captioning is. It rarely matches what's said on screen, and sometimes is just pure gibberish. I feel very sorry for any hearing-impaired people who are trying to understand what's being said on TV with that as their aid.

This is just the beginning of what may be an AI nightmare, since AI does not really think and it can come up with word combinations that are unfortunate.

I spent a few minutes with a friend thing of word combinations that AI might not detect as wrong... such as talking about a fan doing the job of blowing air through a server farm or personal computer.

I see the shortcoming for AI in a live broadcast. But it seems to me, in a scripted program or one with interviews produced in advance, such as "20/20", you'd simply give AI the script and ask it to feed it to program in the closed captioning channel.
 
Closed captioning on live newscasts isn't much better. I can't imagine having to listen to the dialog, remembering exactly what was said and typing it ... all the while, listening to the next words and remembering them.
Sometimes, the closed captioning results can be baffling, making no sense whatsoever.
 
At least in my community, at least one station feeds the script for the prompter to the closed captioning device.

The chatter between segments isn't shown, but the rest of the newscast is captioned accurately.
 
Closed captioning on live newscasts isn't much better. I can't imagine having to listen to the dialog, remembering exactly what was said and typing it ... all the while, listening to the next words and remembering them.
Sometimes, the closed captioning results can be baffling, making no sense whatsoever.
It's been----dear Lord, 11 years---since I worked in TV, but at that time, you could opt for a live closed caption human or a speech-recognition device. The human cost more but, as you note, it's far from foolproof. Court recorders are about the only people who can handle that pace.
 
At least in my community, at least one station feeds the script for the prompter to the closed captioning device.

The chatter between segments isn't shown, but the rest of the newscast is captioned accurately.
Stations in large markets are not permitted to use that method.
 
Stations in large markets are not permitted to use that method.
Due to union rules? Is there even a union for stenographers?

I'm not sure I've ever seen a local TV newscast that didn't simply have the Teleprompter script put into the CC device... but I spend very little time even in the top 50 markets.
 
Due to union rules? Is there even a union for stenographers?

I'm not sure I've ever seen a local TV newscast that didn't simply have the Teleprompter script put into the CC device... but I spend very little time even in the top 50 markets.
At the time, it was an FCC rule.
 
From the FCC Rules:
The major national broadcast television networks (i.e., ABC, CBS, Fox and NBC), affiliates of these networks in the top 25 television markets as defined by Nielsen's Designated Market Areas (DMAs) and national nonbroadcast networks serving at least 50% of all homes subscribing to multichannel video programming services shall not count electronic newsroom captioned programming towards compliance with these rules.
 
A friend of mine's first job in TV was to do the subtitles for the BBC Parliamentary channel, the equivalent of C-SPAN, live subtitling debates. They worked in shifts of just 20 minutes - always in pairs, 20 minutes on, 20 minutes off. Any more than that in a block, and they would find the subtitlers would burn out and errors would creep in, not what you want during a Parliamentary debate.

Interesting that more and more younger people are starting to use subtitles regardless of whether they are hearing impaired or not, so these kinds of errors are going to become visible to a much larger audience:


Apparently, four-fifths of people aged 18-25 say they use subtitles "all or part of the time". I'm not one of them, I'm too old! But I do use subtitles the vast majority of the time. Partly because so many TV dramas nowadays are filmed in what feels like mumble-vision, where characters drawl and mutter their way through scripts, and partly because I find it easier to follow the story with a text prompt as to what's happening.


YouTube has had the facility to auto-generate captions on videos for a long time, presumably powered by Google's AI, but the results are often questionable especially when the presenter has an accent other than "generic mid-Atlantic".
 
Apparently, four-fifths of people aged 18-25 say they use subtitles "all or part of the time". I'm not one of them, I'm too old! But I do use subtitles the vast majority of the time. Partly because so many TV dramas nowadays are filmed in what feels like mumble-vision, where characters drawl and mutter their way through scripts, and partly because I find it easier to follow the story with a text prompt as to what's happening.


I used them when my first wife and I were on different sleep schedules. I spent a few years watching most TV with the sound off and closed captioning on---which is where I discovered just how bad they can be.

Now, a decade later, the CC only comes on if the sound mixing on a TV show or movie is so bad that the dialog is hard to hear. Doesn't happen often. And if it's a movie, subtitles will be accurate.
 
I have ranted for 25 years to anyone who'll listen about how awful closed captioning is. It rarely matches what's said on screen, and sometimes is just pure gibberish. I feel very sorry for any hearing-impaired people who are trying to understand what's being said on TV with that as their aid.

Sometime in the early 2000's I was watching a scripted drama on one of the major networks, with CC turned on. A couple of characters were conversing, and the captioned dialog included an anti-Semitic slur. The actual spoken dialog was completely different and slur-free.

I suppose that the Standards and Practices Department had that word removed from the script before the scene was shot, but the captioner somehow did not get an updated script.
 
From the FCC Rules:
The major national broadcast television networks (i.e., ABC, CBS, Fox and NBC), affiliates of these networks in the top 25 television markets as defined by Nielsen's Designated Market Areas (DMAs) and national nonbroadcast networks serving at least 50% of all homes subscribing to multichannel video programming services shall not count electronic newsroom captioned programming towards compliance with these rules.
Outside of those markets, I suppose one could also self-certify compliance (example)

 
A friend of mine's first job in TV was to do the subtitles for the BBC Parliamentary channel, the equivalent of C-SPAN, live subtitling debates. They worked in shifts of just 20 minutes - always in pairs, 20 minutes on, 20 minutes off. Any more than that in a block, and they would find the subtitlers would burn out and errors would creep in, not what you want during a Parliamentary debate.
I don't know how much this is done on TV these days, but every year, the Methodists from Western North Carolina get together at Lake Junaluska west of Asheville. Several thousand people, about half pastors and about half "lay", attend. There has been someone doing sign language on the stage for as long as I can remember. And that person gets replaced after a few minutes.
 
Due to union rules? Is there even a union for stenographers?

I'm not sure I've ever seen a local TV newscast that didn't simply have the Teleprompter script put into the CC device... but I spend very little time even in the top 50 markets.
I do TV, and you can't use teleprompter text in closed captioning. For one thing, there are notes in the prompter for which anchor should read, breaks in stories, etc., that wouldn't work in closed captioning.
 
Status
This thread has been closed due to inactivity. You can create a new thread to discuss this topic.


Back
Top Bottom