Alien 3 - Audio Description

[NecronomIV]

The proof is in the pudding!

I was really wanting to know if it would work, or whether it would need a re-think. Kyle generously recorded the credits, the first 4:30 of the film, to see how it played. It's dense, but I think it works just fine.

It gives me the confidence to keep going with the rest.

[NecronomIV]

"The ADNA Presents ..." is a podcast that interviews people involved in audio description. If you're interested in my work here, I'd encourage you to listen to even half of this episode:

It's really good. It effectively captures a blind person's experience and feeling of inclusion when watching film and television and even live events via audio description, and the difference between good and bad audio description.

It's the sort of thing that makes me more determined to do the best possibe job!

[NecronomIV]

A little slow going, and I lost a week and a bit due to travel. I'm now 45 minutes in.

It strikes me that I've never actually discussed the how I'm actually doing what I'm doing. I've never posted my set up.

Fundamentally, audio description is someone narrating the movie in between dialogue and important sound, so it follows that writing audio description requires defining a video in-point, an out-point, and the text to be narrated timed to fit that gap.

For example, the creature-birth/funeral happens around the half-hour mark of the Assembly Cut:

00:29:59,130 --> 00:30:03,850
Dillon and the prisoners raise their fists. In the abattoir, the newborn alien opens its razor jaws, and the inner jaw slides out.

(This is an unedited cue. "Razor jaws"? That's what revision is for. Also not spellchecked yet, and it sorely needs it).

What's fantastic about that is that it's functionally the same process as adding subtitles! These two activities map cleanly onto each other. The difference is that with subtitling, you subtitle when people are talking and with audio description, you write when people are not talking.

The upshot of this is you can use subtitle software to write audio description!

And there are loads of free subtitle editors.

I use "subtitld" which has a nice interface, and works on Linux, which is my native ecosystem.



The interface, broken down, is:

• The video on the right
• The time-line and audio wave-form at the bottom, along with the cues.
• On the left is a list of all the cues
• At the bottom, an editing pane for the currently selected cue.

Pretty much you identify a gap in the dialogue, make a new cue, drag it so it doesn't extend over any dialogue or essential sound, and then start typing.

Then, 7,500 words later, boom! You have an Alien 3 script. 

[Kradan]

Thanks for the write-up, @NecronomIV ! Quite insightful and fascinating, also very nice of you to put your time and effort into such a project

[SiL]

@NecronomIV  would it be beneficial to have an SRT of the film to see where dialogue is, so you could more quickly identify where it isn't?

[NecronomIV]

@NecronomIV  would it be beneficial to have an SRT of the film to see where dialogue is, so you could more quickly identify where it isn't?

That is an excellent question from someone who has clearly given it some thought.

Actualy subtitles don't really help, except on occasion to clarify a bit of dialogue or how to name an object or character. However, I suspect it may come in useful when they come to using AI to analyse material and identify gaps and the general intent of a scene.

Now, yes, you could create a tool that does the math and discovers significant gaps and pre-creates spaces to add the narration in, but it would be a very blunt instrument and not sensitive to the soundtrack of the film, and would likely create a lot of spaces that would be un-needed.

I'm not sure AI will ever replace human describers — there's a sort of poetry and brutal efficiency that I suspect an AI just wouldn't get — but if they do, subtitle files will be part of what they feed to it to get it to work I'm sure.

Existing audio description may also form part of a data-set to train AI understand to understand movie visuals and context. Such data-sets already exist, such as MAD.

Great question.

[SiL]

@NecronomIV

Sorry, I don't think I've made myself at all clear.

You said

Pretty much you identify a gap in the dialogue,

What I'm asking is if it would be useful to have an SRT file of just dialogue so you could identify those gaps automatically.

For example, I have transcripts generated in Premiere of just the dialogue in all of the movies. Would such a file help in streamlining identifying those gaps, or is there already a simple system?

[NecronomIV]

@NecronomIV

Sorry, I don't think I've made myself at all clear.

You said

Pretty much you identify a gap in the dialogue,

What I'm asking is if it would be useful to have an SRT file of just dialogue so you could identify those gaps automatically.

For example, I have transcripts generated in Premiere of just the dialogue in all of the movies. Would such a file help in streamlining identifying those gaps, or is there already a simple system?

No, not really. As I said, you could do the math and map all the possible gaps (if, that is, the subtitles were perfectly aligned with the dialogue) but it wouldn't be sensitive to other parts of the soundtrack such as sound effects and important music cues.

The other problem would be that you'd have more of thes gaps than you need - you don't fill up every single gap, that's fatiguing for the listener, like having an annoying person chattering in your ear the whole time ... and you seldom use all of a gap, just what you need (either at the beginning, middle or end).

So in theory yes... ish. The idea is sound. In practice, rather than automating a job, it would give you too much to sift through, and slow you down. But as I said, it's the kind of thing that might be useful input for an AI.

Finding the gaps is the easy bit, and is generally quite an intuitive and organic thing. The tricky bit is the writing, and deciding what to include and what to leave out.