Sophie Muller: Why I’m Seeing More People Turning to Human Subtitles

I’m seeing more and more people clock on that human subtitles are much easier to work with than AI subtitles.

Now, you may be thinking ‘Sophie, are you crazy?’ Well, don’t be so quick to judge because I’m here to explain.

When representing VoiceBox at the IBC, I had a chat with someone who works in the media space  about subtitles, and they just confirmed what I’ve heard quite often from clients about AI subtitles.

Too much time is spent editing AI subtitles! And media companies don’t want their internal staff spending lots of time on this.

You see, with raw AI English subtitles, we’ve found that accuracy rates are around 80%. In theory, that sounds ok. But in reality, you’re looking at around one in five words being wrong. This is no good for your audience trying to follow along.

Whether your audience is deaf and hard of hearing, watching content in a busy/noisy place, a non-native speaker or someone who is neurodivergent, subtitles are very helpful — but only if they are correct.

Subtitles are so helpful in fact, that four in five (80%) 18 – 25-year-old TV viewers said they use subtitles ‘all or part of the time’, Stagetext figures show.

There is demand, particularly with younger generations. And for good reason, plenty of studies show that subtitles improve comprehension, memory and attention around a video.

It would be fair to say that it’s generally accepted that they increase engagement as well, given the wider audience that can be reached in many different situations.

More about AI Subtitles

So, back to this theme of what I’m hearing about AI subtitles.

While they are cost-efficient, as I touched on, no one wants to put out subtitles with only 80% accuracy (at least I think!).

Be mindful that this can differ with audio quality, accent and the language the subtitles are needed in (you would imagine that AI will be better trained in English than Swahili, for example, as its more commonly spoken).

While they are cheaper than human subtitles, people are spending time manually editing them, and that time itself can be costly.

Firstly, if they’re not trained to edit subtitles efficiently, it can be very time-consuming.

And secondly, because their time could be spent doing other things that they are trained to do!

Even for trained professionals, I’ve heard that working on the in-and-out cues, line breaks, segmentation and working out the missing text means that it’s often easier to do it from scratch.

As a result, I’m seeing more people come to us for human subtitles to avoid the manual editing or for AI subtitles with human post-editing, so the job is done for them. That’s something my team at VoiceBox has seen over the last few years too: that it can be quicker to write human subtitles from scratch than edit them.

I think people wanted to try AI subtitles because there was a lot of noise about them, but now that they have, I think more people will stick with fully human or human-in-the-loop subtitles.

What you get with humans

With humans (isn’t it strange that we’ve reached a point in time where we have to say ‘human’ services), you know you’re getting 100% accuracy and the whole process is taken care of for you. You can just focus on the job you’re being paid to do — and the likelihood is that isn’t editing AI subtitles!

If you go for open captions, then they are embedded into your content. If it’s closed captions you opt for, a caption file like an SRT can be delivered.

Don’t forget that humans can provide SDH (Subtitles for Deaf and Hard of Hearing), including details such as sound effects and music cues. AI doesn’t typically do that. And in my book, SDH should be the standard.

With all that being said, I don’t blame clients for turning back to human subtitles. In fact, if you want guaranteed accuracy and to avoid the hassle of editing, I’d actively encourage it!