logo Sign In

CatBus

User Group
Members
Join date
18-Aug-2011
Last activity
26-Jun-2025
Posts
5,996

Post History

Post
#1541231
Topic
Project Threepio (Star Wars OOT subtitles)
Time

Rondan said:

CatBus said:

“Romani ite domum”

Hahaha
Great choice for a name.
Did you have to write it a hundred times too?

I must have seen this movie at least a hundred times. Almost as many times that I have seen Star Wars.

Many of my codenames are plays on words, plays on numbers, references to pop culture and current events, or all three. And I’m a big Python fan. When I’m secretly worried that my machine translation is going to come up with bad translations, that anxiety comes out in the codename 😉

Speaking of which, version 14.0 was a bit rushed, to make it out in time for Despecialized ROTJ 3.0. I had plans for a lot more machine-translated languages, but those had to be deferred, and will show up in a future release.

The bigger news is that while I was initially conservative with machine translation, I am at least initially getting good results with languages that I considered longshots. Specifically, Georgian. Georgian isn’t really even somewhat related to any other major language, so I picked Russian as the starting language – simply because some Russian colloquialisms and other terms may have worked their way into the language, and “problematic machine translations from Russian” may be the sort of strangeness that Georgian speakers may already know how to accommodate.

How do I measure if the translation is any good? Well, if the translation round-trips from Russian to Georgian, and then back again to English, and it’s still intelligible, I figure it’s probably okay. In this case, it’s not just intelligible, but usually pretty solid (and sometimes merely intelligible too). So assuming this keeps up, I may be able to add some very interesting languages.

Sadly, it appears that it may be more the quality of the machine translator for that particular language than any proximity to an existing translation, so I just won’t know until I try, and I may not be able to predict which languages will work. For example, Faroese is very close to Icelandic, and that attempt failed badly in the early stages.

Post
#1540890
Topic
Harmy's RETURN OF THE JEDI Despecialized Edition HD - V3.1
Time

More possible fixes:

I’m still noticing a grading shift mid-wipe at 1:21:38, but I’m not sure if it’s theatrical or not.

Okay, I know the Boba Fett flyby is supposed to look janky when you watch it frame-by-frame, but could you check against 4K83 to see if it looks different there? Something about the current flyby looks like very low res video scaled way too large, and I recall this was your only GOUT-sourced despecialization in 2.5.

Post
#1540888
Topic
Harmy's RETURN OF THE JEDI Despecialized Edition HD - V3.1
Time

On the bug report front –

The 1080p encode seems to be missing a framerate in the H264 file. This works fine in the MKV, but if you try to demux/remux, you may have to manually set the framerate or you could get issues.

The 2160p encode has no black bars – not a problem for most people, but it does cause problems when using PGS subtitles. PGS subtitles generally need a 16x9 frame or weird and bad stuff can happen during playback.

Post
#1540455
Topic
Project Threepio (Star Wars OOT subtitles)
Time

Project files have been updated to version 14.0 (codename: “Romani ite domum”), and the first post has been updated. Please PM me for temporary download links until the files are available at some more permanent locations.

Rough summary of changes from 13.0 to 14.0:

  • Project Threepio now supports machine-translated full subtitles. At least for now, I am being extremely conservative about using machine translation. Machine-translated subtitles will be clearly labeled for those who want to avoid them, and at least for now, I am limiting its use to translations between closely-related languages to cut down on translation problems with idioms and sci-fi vocabulary. However, even in the worst-case scenario, we may still attract the attention of people who could improve the subtitles.
  • Canadian French subtitles are now trilogy-complete, using machine translations for the missing films
  • Bosnian subtitles have been added, using a fansub for ESB and machine translations for the other films
  • Added Tajik subtitles, using machine translation, and also redid Afghan Persian (Dari) subtitles using machine translation, since the existing ones were not good
  • Added Telugu and Tamil titles-only subtitles (machine-translated) to accompany ROTJ dubs
  • Galician titles-only subtitles updated to accompany new, more complete, dubs
  • The render script now works even when you do not have any of Project Threepio’s TTF fonts installed (OTF fonts must still be installed to render subtitles for those languages), as long as you’re working from the complete Project Threepio project folder. This should help with scenarios where the system has different versions of the fonts installed than Project Threepio needs. The new script allows you to specify how many threads it should use, to improve performance and stability across platforms.
  • Improvements to Spanish translations for Star Wars (thanks to CAL0901 and laozi)
  • Improved Dutch subtitles for ESB (thanks to frater)
Post
#1535063
Topic
Project Threepio (Star Wars OOT subtitles)
Time

It’s been a while since I’ve posted one of my technical progress reports, so here goes.

This project has been through a few text renderers in its day. First, we had an entirely manual Windows-only renderer (easySUP), then a scriptable and mostly-but-not-quite cross-platform one (Pango). Currently, the project is using Chrome – or to be more accurate, Chromium, the open-source project behind Chrome and other Chrome-like web browsers. This has solved a lot of problems. It’s scriptable, it’s truly cross-platform, and it easily deals with the complicated technical requirements behind rendering some of our non-Latin scripts.

But there’s always something, and I’ll just come out and say it. I hate the way Chrome renders text on Windows. Firefox does way better at this. I hated Chrome’s text rendering even as I moved the project to Chrome, hoping it would someday get better. There are lots of bugs filed on this, all lumped under a single uber-bug here. The gist is this: Chrome’s text is too light, too thin, too spindly. Bold text looks more like semibold, semibold looks more like medium, medium looks more like regular, and so on. That doesn’t mean that text is bad or unreadable or even unpleasant (some people prefer it this way, and that’s okay), but it does mean that the appearance of the text differs from the intent of the person who chose that font and weight. And for subtitles, that’s important.

Font weight comparison

With subtitles, you want your text a little on the thick side, so they don’t get lost against a busy, moving background. Some people use bold subtitles, but that’s a little hardcore for my tastes – I prefer to ease it back one step. In my opinion, semibold is the ideal font weight for subtitles. Thick enough to really stand out, but not so thick it draws too much attention to itself. But, for the past few releases, Project Threepio’s subtitles have been a touch lighter than that, and I just tolerated it because Chrome solved so many other problems for me. And the difference is honestly subtle – most people couldn’t distinguish Medium from SemiBold, and I knew this was the sort of thing that was likely bothering only me. I could have switched everything to Bold, but the result was still slightly thicker than SemiBold on Windows, and it’d screw up non-Windows platforms, where Bold is Bold and SemiBold is SemiBold.

But if you follow that link to the Chromium bug report, you’ll see that progress has been glacial. This does not seem to be a Chromium project priority. Edge implemented a fix called enhanced-text-contrast and it works as advertised. But not only is that fix unavailable on other Chromium-based browsers, it’s still experimental on Edge and turned off by default. The default state is very important to me, because when running in headless scripted mode, you can’t always configure esoteric options – you mostly just run with the defaults.

So, while researching my options, I stumbled across an old fork of Chromium, called GDIChromium. The reasons for creating this fork were somewhat different than what I was looking for, but the resulting text is definitely what I wanted. The fork is old and doesn’t look maintained, so I would definitely not recommend it for day-to-day browsing, but it’s new enough that the headless scripting works. I’ve tried it out, and things look very good. Ideally Chromium will get around to fixing its text rendering on Windows, but until then, I think I’ve found a fine stopgap solution.

So, to make a long story short, I’m re-working the subtitle rendering script yet again, and the next release will likely have very slightly, perhaps imperceptibly to most, thicker subtitles.