Live Teletext Subtitles

I don't have hearing difficulties, but I do use subtitles to help improve my spelling. I can't work out whether live broadcasts use very fast typists or voice recognition software to generate the subtitles.

On one hand the live subtitles place emphasis very well. For example, exclamation marks are used in the right places, and you sometimes get an ellipsis followed by a question mark (...?) after an incomplete question, as opposed to just a question mark on its own. How would a voice recognition program know how to do this? The subtitles also handle context reasonably well. For instance on yesterday's BBC Breakfast, the presenters were talking about Wispa chocolate bars and the correct word 'Wispa' appeared in the subtitles. A computer couldn't possibly know whether they were referring to a brand name or the word 'whisper'. Taking all of these factors into consideration, you'd think it was a person typing the subtitles.

However, there are too many mistakes. Gerry McCann suddenly turns into a 'jerry can', for example. A typist would have to be very stupid to make that kind of mistake, and would probably be sacked for making the overall amount of mistakes that do appear. Some complicated words are often omitted, particularly place names. So '16 police have been killed in China's Xinjiang province' appears on screen as '16 police have been killed in China'. This makes me think it's a computer because it doesn't recognise the name of the province, and therefore decides not to include it in the subs at all. Then again, a typist wouldn't have the time to look up the spelling either.

It's very confusing. The only thing I can gather for certain is that presenter's and reporter's voices are not recorded for any recognition program to learn, because the accuracy and style of the subtitles is consistent between presenters and guests.

So does anybody have a definitive answer as to how the live subtitles are actually produced? Also, I wouldn't mind working as a subtitler for pre-recorded programmes. What kind of qualifications are required for that?

It's my understanding that the system is manned by live personnel, but that the words are not typed out longhand -- there is a shorthand system in place that allows operators to enter words as they would be said (dunno how this works), enabling a much larger number of words per minute to be entered. A computer then translates these into English -- with the sometimes strange results you see.

The alternative would be that the operators would not be able to keep up.

There are two systems in use for live subtitles, AIUI.

The longest established is using stenography, similar to the way court recorders transcribe the court proceedings, using a simplified keyboard in which each key represents a character used in shorthand - see http://www.bivr.org.uk

The newer method involved speech recognition, but not of the broadcast itself. Speec recognition needs to be trained to understand the way an individual speaks, so somebody re-speaks the material, perhaps with the aid of a script if one is available, into the system.

I applied for a job as a trainee subtitler a while back - one of the job requirements was apparently "to be able to speak and listen at the same time"

This leads me to believe that it is a Ministry of Truth-style set up, with the program being piped into your ears, and you speaking into a speakwrite.

So if you're talking about Wispas, the ration is going up from 20g to 24g per week.

There is another type - where the subtitler has the text all ready to go in the machine and then just feeds it out to match the broadcast.

This happens when there is a script but they haven't had a chance to make it into proper subtitles yet. It's also common with news stories (when the reporter has put his/her script into the running order) and when a programme is re-edited and they haven't had a chance to amend the subtitle file

Taking a dump of the rundown from ENPS and feeding it into the subtitle system is the usual work around if you get enough warning your live subtitler is unavailable. The regions do not have their own dedicated staff to produce the live titles based at the studios, they have people dial in from elsewhere instead - which goes wrong every so often.

And sods law dictates that should you get prior warning and load the system up from ENPS, the news director will then change the running order after you have run the export and you put out subtits that are somewhat at odds with the audio.

As others have mentioned, 're-speaking' is used increasingly more for subtitling of live programmes; it's far cheaper to train a re-speaker than a stenographer.
The subtitler listens to the programme, and then repeats the dialogue, including punctuation.
Often a researcher will have reviewed the script or running order beforehand, and prepared a word list of complicated words, which the subtitler will use to train the speech recognition software (often just IBM ViaVoice).

Red Bee Media have the contract for BBC and Channel 4, amongst others. They take care of the BBC's regional news too, often remotely. I'm not sure if they still do it, but for a while, overnight subs for News 24 were originated in Sydney - Red Bee own the Australian Caption Centre.

And, since you ask, they're currently hiring: http://www.redbeemedia.com/html/current-vacancies.html
(vacancies pop up relatively frequently on that page for posts in London and Glasgow)

Thanks for the information people