The Perils of Automatic Subtitling

16/05/2016 15:23 | Updated 17 May 2017

My production company Jinx Media is a tiny independent. We never have enough money or time. Somehow we've survived for thirteen wonderful years of making low-budget horror movies, but it's never been easy.

In order to make sure our movies get seen, we're always on the look-out for new developments in the industry. Last week's launch by Amazon of Amazon Video Direct prompted a minor scramble in the Jinx offices to see how quickly we could prep back-catalogue movies to take advantage of the new platform. The first thing we noticed was that all videos uploaded to Amazon's new service require closed-captions.

This is extremely sensible and forward-thinking, and I approve wholeheartedly. It did, however, present an immediate challenge in terms of getting our back-catalogue of movies onto the service, since none of them had closed-captions already prepared. After a bit of asking around for quick solutions (our budget wouldn't unfortunately allow for the sensible option of getting professionally prepared subtitles put together by a third party), a colleague suggested that the best starting point would be to use a free online service to prepare 'auto-detect' subtitles, which could then be proof-read and edited before being exported and attached to the new version of the movie for Amazon.

This seemed like a damn good starting point. I didn't even know such a thing existed, although I've fooled around with voice recognition software before when getting large amounts of text written in a short period of time. Our test movie was Hellbride, which seemed a sensible enough choice as we'd only just re-released it onto VOD the week previously, and therefore had all the other deliverable materials ready to rock.

And so, to the automatic subtitling.

It worked!

Well, it sort of worked.


Hmm. Well, no system is flawless. I'd already been led to expect that I was going to have to do some fairly serious proof-reading and editing. And surely the bit about the dog would just be a one-off, and the vast majority of...


Ah. Right. At this point, I started scrolling through the timeline to see if I could find a single caption that was actually on the money. But the gibberish just kept coming and coming.



Ah, yes. Coffee coffee coffee. I'm going to need it.

Because I think it's going to be a long night.