Parallel colaborative fansubbing proposal

[ScR3WiEuS]

jfs, mendoi implemented a worktracker a while ago. a fairly easy one with a php bot. most members were just flaming or ignoring it. subbers are stubborn creatures, they don't like to change habits :x

[lamer_de]

If you really need that kind of parallel work setup, I don't see why it shouldn't possible with Aegisub as it is now plus the use of IM/IRC.

The translator pastes lines as soon as he finished them on IRC, timer copy/pastes them into aegi and times them. At the same time, the editor does paste them into aegi as well, generating untimed/fake timed lines and edits them. If you need/want, you can stick TC in between. Thanks to the paste-over feature in Aegi, a merge of the correct timed and the correct edited script can be done in seconds by the timer (if he split lines during the process, he can split the edited script quickly, too). Sign TS and encoding are disconnected from the other tasks anyway, and dialogue styling can be done in 3mins if the script has the correct hints (like "<char name> thought" in the actor field etc).

[dj_tjerk]

copy paste gives me RSI >_>

Also.. i often have to work at different pc's.. it'd be very easy if i could just press a button in aegisub which'll download the needed files, instead of having to use ftp every time; makes me go mad.

[ArchMageZeratuL]

Originally Posted by lamer_de

If you really need that kind of parallel work setup, I don't see why it shouldn't possible with Aegisub as it is now plus the use of IM/IRC.

The translator pastes lines as soon as he finished them on IRC, timer copy/pastes them into aegi and times them. At the same time, the editor does paste them into aegi as well, generating untimed/fake timed lines and edits them. If you need/want, you can stick TC in between. Thanks to the paste-over feature in Aegi, a merge of the correct timed and the correct edited script can be done in seconds by the timer (if he split lines during the process, he can split the edited script quickly, too). Sign TS and encoding are disconnected from the other tasks anyway, and dialogue styling can be done in 3mins if the script has the correct hints (like "<char name> thought" in the actor field etc).

Because that would be a lot of work and prone to error... I want a system where nobody will have any more work than they would in a normal process.

[Tofusensei]

Originally Posted by Quarkboy

I've actually done some serious thinking about what the true minimum time for speedsubbing would be, and I believe it is perfectly achievable BY ONESELF.

Here's the flow:

Step 1: Live in Japan, own a HD hard drive recorder.
Step 2: Watch the show as it airs, while recording it. Take translation notes if needed, look up things during commercial break. (30 minutes)
Step 3: Set up raw encode from HD capture <- these can be extremely high quality with good equipment, since some simply capture the mpeg stream from the digital broadcast... at least that's what mine does (although I haven't figured out how to get around the copy encryption)).
Step 4: During h264 encoding, translate, time and edit the show. Time as you translate for speed's sake. Do an edit/timing check pass.
Step 5: Since the raw encoding takes about 2-3 hours for 1280x720 with a medium speed Core2duo and medium settings, you have plenty of time to fix up the scripts, which can be timed based on the original mpeg capture file. Don't worry about fine-timing since we'll be softsubbing anyway.

When the raw encode is done, mux in your script and create and seed a torrent somewhere.

Total time from airing, ~4 hours, less if you release SD and can translate really fast and/or use xvid instead of h264 (the fastest I've ever translated a show was 45 minutes... but I'm not the fastest out there. I average more like 2 hours).

I believe that that's really the theoretical limit, since encoding from the tv capture is always going to have to be done (either by you or the capper), and that takes longer than translation, timing, typesetting, editing, and qcing, if done as fast as theoretically possible.

Not that I recommend anyone try this... It's possible, but I can't imagine it would make a particularly good sub.

FYI, I am someone with good experience in this area. We've done what you described in approximately 8 hours before, 2 hours for pre-distro and distro, 10 hours from the moment it airs until final release. That was a very high quality sub as well. That is the all-time speed record for myself.

1) Raw captured in Japan, encoded on the fly, had on my PC 10 minutes after it finished airing. I had access to a streaming copy as well but a translator will not really have any benefit from watching it live other than getting a better understanding of the episode in full. (00:30)

2) Translated, 2 hrs (02:30)

3) Timed, 45 minutes (03:15)

4) Concurrent editing, typesetting, pre-filtering of downloaded raw, 2 hours (05:15)

5) QC, 90 minutes (06:45)

6) final encode, 45 minutes (07:30)

7) upload, pre-distro and release, 2 hrs 30 minutes (10:00)

10 hr release from airing with all the steps in there. This was for one of the DB/L-E Tsubasa Chronicle episodes. I think DB might have done a Naruto release faster than this.

So yes, your 4 hour theoretical peak (assuming everyone has uber fast fiber optic connections) is probably accurate.

That being said, I do not see a single step that this endeavor would improve this process by more than maybe 1 hour total. This is why I initially said it is impractical (impossible) in practice. It's a solution to a problem that doesn't really exist.

-Tofu

[ArchMageZeratuL]

Originally Posted by Tofusensei

That being said, I do not see a single step that this endeavor would improve this process by more than maybe 1 hour total. This is why I initially said it is impractical (impossible) in practice. It's a solution to a problem that doesn't really exist.

Of course, I'm not talking about THAT workflow.

Consider the alternative:

1. Download raw from a torrent

2. Translate, Time, Typeset, Edit, QC

3. Remux with raw and release


With THAT workflow, parallelization would make it many times faster.

[Nicholi]

Even if it doesn't make it faster, it seems like it would be a favorable way of working on projects (even where speed does not matter). You can see the history of every line, know exactly who made/changed each line, what lines need to be worked on or checked by the relevant parties, all the information you need is at your fingertips without asking anyone. Most significant is, everyone does not even need to be present at the same time. Everyone could individually work on the script as they can and eventually it will be done.

It sounds like it would be a perfect system for fansubbers in completely different timezones. In theory they would never have to talk to each other "face to face" ever. Just leave notes to each other and work on the script as required. Damndest efficiency I've ever heard of.

Resist it all you want fansubbers of ye olde time. If it can be done, it's not slowing you down any...since you won't be using it, lolz. Cheers to anyone else who trends into the madness realm of efficient subbing.

[False Dawn]

Originally Posted by Tofusensei

5) 90 minutes

Is this a usual step of the process that I was not hitherto aware of?

[Tofusensei]

Originally Posted by False Dawn

Is this a usual step of the process that I was not hitherto aware of?

lol, thanks ^^;

I fixed that to say QC.

-Tofu

[Tofusensei]

Originally Posted by Nicholi

Even if it doesn't make it faster, it seems like it would be a favorable way of working on projects (even where speed does not matter). You can see the history of every line, know exactly who made/changed each line, what lines need to be worked on or checked by the relevant parties, all the information you need is at your fingertips without asking anyone. Most significant is, everyone does not even need to be present at the same time. Everyone could individually work on the script as they can and eventually it will be done.

It sounds like it would be a perfect system for fansubbers in completely different timezones. In theory they would never have to talk to each other "face to face" ever. Just leave notes to each other and work on the script as required. Damndest efficiency I've ever heard of.

How do you treat lines that a translator leaves to go back to later? This happens all the time.

How do you treat when the translator goes back to change something that is made apparent by something that happens later in the episode? This happens all the time.

Typesetting is already started concurrently with the translation in most speedsubbing groups so there's no real benefit there.

Why on earth would you begin QC until the script is finalized? That's the dumbest thing I've heard this entire thread.

I've been in a physical room with an entire fansubbing team before and did EXACTLY what you are describing and it provided very little actual speed benefit (1 hour or less).

Originally Posted by Nicholi

Resist it all you want fansubbers of ye olde time. If it can be done, it's not slowing you down any...since you won't be using it, lolz. Cheers to anyone else who trends into the madness realm of efficient subbing.

I think I've treaded much further into the realm of efficient subbing than you ever will, but now I'm just trolling

-Tofu

[Tofusensei]

Originally Posted by ArchMageZeratuL

Of course, I'm not talking about THAT workflow.

Consider the alternative:

1. Download raw from a torrent

2. Translate, Time, Typeset, Edit, QC

3. Remux with raw and release


With THAT workflow, parallelization would make it many times faster.

Many times faster? So we're talking 10 hours now... Many times faster would put it at what... 3 hours?

-Tofu

[juggen]

Originally Posted by Tofusensei

3) Timed, 45 minutes (03:15)

-Tofu

Who the hell can time in 45min...? O_O

[Tofusensei]

Originally Posted by juggen

Who the hell can time in 45min...? O_O

I know of 3 timers who can time an episode in 30 minutes or less, I've seen them do it. I have no idea how.

One of them says he chain-smokes while he does it

The fastest timers all use SSA, by the way.

-Tofu

[juggen]

lol, I'm silverfire's bitch >_<

1 [10:11:56] <@silverfire> im too fucking lazy to log into asuki so post IRC logs in reply to tofusensei for me or sth
2 [10:12:05] <@silverfire> dialogue: <20min
3 [10:12:17] <@silverfire> songs: <15min (90sec song)
4 [10:12:28] <@silverfire> aegisub, spectrum analyzer
5 [10:12:35] <@silverfire> mouse5 = play line to end
6 [10:12:41] <@silverfire> mouse4 = set line and play to end
7
8 oh and mouse5/mouse4, etc = custom macros coded in autohotkey

1 [10:19:48] <@silverfire> add something saying that
2 [10:19:50] <@silverfire> i dont time by ear
3 [10:19:53] <@silverfire> i time primarily by eye
4 [10:20:01] <@silverfire> and the key is to learn how syllables LOOK.
5 [10:20:12] <@silverfire> because then you recognize sentences by how they look
6 [10:20:16] <@silverfire> and frequency ranges

[Tofusensei]

Originally Posted by juggen

lol, I'm silverfire's bitch >_<

The way he describes it is exactly right. There was one timer from AnimeOne who used to time in less than the length of the episode, assuming it wasn't riddled with background songs. I don't know about other software but its a matter of:

1) Look at waveform
2) Left click, right click, click grab times

Move on.

Once you get good and don't need to double-check your work much, it's not that hard to do it quickly. I felt 45 minutes was a reasonable estimate.

-Tofu

[sangofe]

The problem with your system, is at least for me, who's a timer, that i often go back and change lines here and there, and always do a check before uploading.
I believe others do the same in other positions as well.

[sangofe]

Originally Posted by Tofusensei

The way he describes it is exactly right. There was one timer from AnimeOne who used to time in less than the length of the episode, assuming it wasn't riddled with background songs. I don't know about other software but its a matter of:

1) Look at waveform
2) Left click, right click, click grab times

Move on.

Once you get good and don't need to double-check your work much, it's not that hard to do it quickly. I felt 45 minutes was a reasonable estimate.

-Tofu

'looking at the right waves, having one waveform only, scrolling with both mouse and keyboard at same time, and at least but not last HOTKEYS, and you should be able to go fast.

[ArchMageZeratuL]

Originally Posted by Tofusensei

I don't know about other software but its a matter of:

1) Look at waveform
2) Left click, right click, click grab times

Well, Silverfire said that he doesn't look at the waveform - he looks at the spectrum. The spectrum is much better, once you are used to it, provided that your computer can take it.

[martino]

Originally Posted by ArchMageZeratuL

Well, Silverfire said that he doesn't look at the waveform - he looks at the spectrum. The spectrum is much better, once you are used to it, provided that your computer can take it.

That's how I time myself as well, I personally don't like the waveform. And as silverfire said (from juggen's post), once you learn how the syllables look on the spectrum it is a rather quick job since you don't even have to rely on the audio itself, but timing from the spectrum just by looking at it is pretty fast and accurate (given that you know WHAT to look for). If I had to give a time comparison, personally I can time roughly twice as fast using the spectrum than the waveform, but then I'm not someone that'd call himself a "timer" and not like I have much experience with the latter either (also I guess this depends on the person timing as well, everyone has their own style and way of doing things). Anyway, just my 2c, kind of...

[sangofe]

Originally Posted by ArchMageZeratuL

Well, Silverfire said that he doesn't look at the waveform - he looks at the spectrum. The spectrum is much better, once you are used to it, provided that your computer can take it.

What is the "spectrum", and how do I "see" it?