- Post
- #933558
- Topic
- Project Threepio (Star Wars OOT subtitles)
- Link
- https://originaltrilogy.com/post/id/933558/action/topic#933558
- Time
PM sent.
PM sent.
If he has a commission from the Queen to do so, then technically he’s a privateer.
PM sent.
Defiantly gonna have a Star Wars marathon
Fight the power.
As for the initial question asked – no, these subtitles don’t line up with the dubs timing-wise, and also they are different translations, so they don’t match even though they’re the same language.
Project Threepio subtitles try to use the best translations of the spoken English lines, without many other considerations. Dubs try to choose words that match lip movements, and other stylistic things that aren’t purely translation–what you really end up with is an entirely different performance, maybe good, maybe not. Star Wars is even more complicated than many movies, because so many lines are spoken by people wearing masks that dubbers often can choose to time their lines however they like. And some dubs, like the Thai dub for example, time the dubbed lines very haphazardly even when there are lip movements they could line them up with.
But, back to the flamewar-worthy part of this, here’s something I actually like that most people hate: Soviet bloc-style voiceover dubs. Never heard one? Try the Polish, Ukrainian, and Russian voiceovers. Basically the original performance is kept intact, and a translator who tries to stay very neutral just speaks over them. It sounds a lot like a speech at the UN or a radio interview in another language. Okay, it’s weird, I admit, but the original performance is left alone, and the voiceover style likely gives very little temptation for embellishment or alteration, which are both significant advantages over regular dubs in my book.
EDIT: Not to mention that the dubs that voice over the crawl use the 1981 text, while the subtitles use the 1977 text, and despecialized dubs have altered SE lines, while the subs use the original lines.
Someone left the cake out in the rain.
Don’t blame it all on Lucasfilm. Plenty of people don’t see any significant visual difference between DVD and Blu-ray, and assume all content on a DVD is more-or-less “DVD quality” and by extension not significantly different than Blu-ray. Therefore, because the GOUT is on a DVD, it has essentially the same visual quality as the Blu-rays. QED, or something like that. You see this all the time, it’s just particularly galling because the GOUT is so awful (which you can blame on Lucasfilm).
Well, you know, it’s the Wall Street Journal. Now even people on dialup know about Harmy. Word spreads slower there.
Well, I’ve gotten further in this particular stab at rewriting the subtitle rendering routine than I ever have before. The main sticking point seems to be that there’s a Windows-specific bug in ImageMagick+Pango, which means even if everything else is perfect (and we’re not quite there yet either), we’ll need to maintain two different rendering engines, which produce slightly different results, which is relevant in the sense that the SRT timings are optimized for the Windows method (the new method can occasionally be one frame off), etc, etc.
And, well, the script is slow as molasses. But it’s slow across all platforms, so there’s that.
Keep in mind, bulb matching is a thing. This is a great way to get the color on the print, but the color on the screen is the color on the print combined with whatever color bias the projector bulb has. So for a seventies bulb, we’re probably talking a little yellower. Not to diminish the importance of this in any way–this will be extraordinarily helpful, but maybe not the final word.
From what I understand, this could relatively easily be incorporated into DrDre’s algorithm. You would need to describe the light used to scan a particular print and the light emitted by a bulb what was in common use when said print had its run in theaters. Of course the accuracy of the final result would depend on how accurate data you fed to the algorithm.
It’s not that it can’t be done, but then you lose the “objectivity” which is part of the draw of DrDre’s method. Choosing the bulb LUT is both subjective (which bulb to choose?) and prone to error. I like what he’s doing, but I see this as more of an initial color correction than a final one, which is really where my comment was headed. You need the objective color correction as a base correction – in fact it’s critical for it to be as objective as possible, and this seems to be doing that – but that doesn’t eliminate the need for subjective corrections afterward.
Keep in mind, bulb matching is a thing. This is a great way to get the color on the print, but the color on the screen is the color on the print combined with whatever color bias the projector bulb has. So for a seventies bulb, we’re probably talking a little yellower. Not to diminish the importance of this in any way–this will be extraordinarily helpful, but maybe not the final word.
Mono has it. They just went nuts with it on the SE track. You’re not going to find anything that sounds like the SE track until the SE. The original mono track effect was much subtler.
Yes, contrary to popular belief, size does matter.
“Matters size does”?
There should be a DVD downscale which fits on a DVD5. It’s not HD, but HD with that much compression would just be trading resolution for compression artifacts.
Re: patents, I suspect he’s not the first person to create such an algorithm, he’s just the first person to do it (presumably) for free, and also while not working for some corporate behemoth who looks at the natural, unfaded results and says, “That looks okay for a start, now let’s teal and orange that puppy!”
If he used the isoscore from me, I got it from Laserschwert. It’s a hot potato.
Well, to be honest, the subtitle images in my project for ROTJ come straight from Harmy, so I didn’t exactly do much work there. But the Star Wars subtitle images are my own doing (and part of why they’re not quite as nice as the ROTJ subtitles), based on lower-quality film scans, with lots of processing in Photoshop to make them more usable.
Oh absolutely. The first clause of “It’s silly to even care about this and here’s my considered opinion on this and why I think I’m right” is hard to dispute, the second is hard not to giggle at.
The fact that nobody cares or even knows that the name got changed is kind of a given, and part of why it took Lucasfilm a couple decades to bother changing the primary name on the packaging.
And I seriously can’t imagine anyone in real life getting fussy about word choice.
My point is, you just did.
If anyone’s interested, here’s what I’m currently working on. Most of the tools in this project rely on some mixture of Perl, ImageMagick and a Java subtitle utility called BDSup2Sub. They work great, are pretty flexible, cross-platform, and all that. The fussiest one of the bunch is BDSup2Sub, and its minor shortcomings are pretty easy to work around.
However, there’s an old legacy bit underpinning this project’s entire subtitle rendering routine. It currently uses a long-discontinued utility called easySUP, using a customized long-deprecated AviSynth subtitling DLL called vsfilter. It’s Windows-only, and it has lots of pretty serious shortcomings I’ve been working around by hand, such as terrible RTL language support, reported problems with Indic scripts, not a lot of formatting flexibility, etc. Most of all, I currently do the initial render for each subtitle by hand, which wastes time and risks errors.
Now, I probably could bypass easySUP and script directly against AviSynth, but since I don’t want to prolong my troubles with the existing system, I’m now writing a Perl script that renders subtitles using ImageMagick+Pango. So far, the results are promising. And, more importantly, if this all works, I can throw out tons of kludgy workarounds for RTL scripts, render subtitles while I sleep, and so on.
All of this will take a while.
There’s no reason to keep going with 9.0. And thanks!
Who am I to turn down a free offer? PM sent.
That’s correct, no changes to Arabic since 9.0, maybe some minor timing things but I can take care of that.
The original subtitles were burned-in to the picture, so it’s unlikely that a file of just the subtitles is out there.
…not that we can’t try. The latest Project Threepio includes graphical subtitle files derived from 35mm scans. Not perfect, but not far off the mark.
It didn’t fit the thread, but it fit the pattern I’d identified for the other two films, and the other two films definitely fit the thread.