Translating what’s said on live television into readable text – it sounds simple enough.

But take a moment to consider the challenges of providing live subtitling across multiple channels, with breaking news, regional broadcasts and over-running sporting contests, then add to that mix the understandably high standards of timeliness and accuracy expected by viewers, lobby groups and Ofcom alike and it might start to seem more complex.

We’ve talked before about using voice recognition software, improving the process of live subtitling and the techniques employed to attain the best outcomes possible – but what you may not know is that we, and the industry as a whole, are always looking at ways to address the challenges we face with live subtitling through ongoing investment in innovation and technology.

THE VOLUME CHALLENGE
As mentioned, a huge challenge presented by live subtitling in a modern regulatory environment comes from its sheer scale, with multiple regional opt outs for some programmes, 24 hours a day of live output for news channels, over 150 daily hours of live output, and a workforce of subtitlers spread across many continents and also working from home.

THE LOGISTICS CHALLENGE
Traditional subtitling software involved little in the way of schedule management and it would sometimes happen that subtitlers were not correctly present for some programmes or accidentally sent subtitles to the wrong channel. Handovers due to shift changes, regional opt-outs or breaks have traditionally been complicated to arrange without the loss or duplication of some subtitles, particularly where the subtitlers are not in the same location.

THE ACCURACY CHALLENGE
Most truly live subtitling is generated using voice recognition software, most commonly Dragon NaturallySpeaking. There are huge benefits to this approach. When coupled with the respeaking method, whereby a subtitler repeats everything spoken on screen in a clear and level voice, with punctuation and colour commands, the recognition and output is broadly excellent. Furthermore, good speech engines are easy to use, so it is relatively easy to recruit and train people to subtitle well. However, historically, there have been some downsides.

The best voice recognition engines avoid releasing text until they receive enough context to be sure of what is being said. This can lead to a significant gap of output on air, followed by an instantaneous release of a large chunk of text. Slowing this glut of text down to a readable speed, as most software does, leads to desynchronisation of subtitles and video and the need to edit or omit sections to catch up. Regardless of the amount of context, voice recognition software will always struggle to understand some of the time, particularly with unusual terminology or names – the kind of vocabulary commonplace in live news. Traditional subtitling software uses housestyles to automatically replace common typographical errors, such as “empire state-building” for “Empire State Building” and the program’s comprehension errors, such as “Nicholas so cosy” for “Nicolas Sarkozy”.

Subtitlers also define macros for alternate forms of sound-alike words, so they can tell viewers whether it is chilly out, or they’re making a chilli, or the sports event is happening in Chile. But macros and housestyles are set up prior to going on air, and are therefore of little use when a new name emerges in a breaking news story.

MEETING THE CHALLENGES
Over the last three years or so, we’ve been looking at how we can improve the quality of live subtitling. We believe that technology and innovation will be key drivers to achieving this goal and as such, we have been investing heavily in building a bespoke platform and software that we believe will help to address some of the challenges listed above. Unique functionality such as the ability to integrate with broadcaster schedules and a respeaking interface designed to be the fastest on the market are just some of the improvements we’ve been focused on.

Live subtitling will continue to be a complicated, imperfect and expensive process for some time to come, but with our new software, Subito, we believe we’ll be able to deliver greater quality to the audience, and, as we’ll talk about in another blog in future, begin to add extra utility and lower costs for broadcasters. We know there will always be more that can be done, and we remain committed to both integrating current technological advancements and driving the next set forward. We will continue to do so until we reach a day when live subtitles are barely distinguishable from prepared ones.

Hewson Maxwell, IT Manager (Spain), Access Services.