Getting Into Fansubbing

[TheFluff]

Originally Posted by kagetsu

Don't be afraid either of tweaking with the 1000's of a second as thats where most of tweaking was done to really hit the sweet spot.

Uh. No. Tweaking in the +10/-10 millisecond range is pointless. One frame at 24fps lasts for about 42 milliseconds, being any more precise than that (+/-4 100ths of a second) never ever matters since all sensible subtitle renderers needs a new frame to draw a new sub line.

Originally Posted by kagetsu

No sound wave visualization but I never liked that anyways really with some other programs, I just needed to see the video.

Audio timing on the video sounds rather imprecise to me...

[pichu]

That reminds me of one karaoke timing (\k) I did, which has over 40 syllables/romaji (I refuse to combine them) within 6 seconds. And, it gets mistimed because of the 'centiseconds' limitation of ssa, but if it were to extend to milliseconds, it won't get mistimed so easily. Remember, this is also known as truncation errors (small errors add up) numerically. It's probably because of medusa, which I used a couple of years ago.

[Sylf]

I really doubt you need that kind of accuracy, actually. Yes, you may be frustrated with the inaccuracy you see on the wave form. But what counts the most is if the karaoke looks in synch with audio on the screen or not. What I've noticed is that just because I have all syllables timed accurately on the wav form doesn't necessary mean it looks tight and accurate on the screen. Psycho-visual effect, I guess?

[Lynn]

hi there

well, I hope this is the correct thread, but it seems quite good for me^^

I'm planning to start some active fansub work now or in the future and would like to work as a translator some day. The problem is I'm not good enough yet, because I started japanese lessons a 3/4 year ago.

So I would like to do some other job to get into this thing. First question is, are there groups out there that teach the newbies? Well I guess I can offer a few useful skills because I did some AMV's with Premiere Pro and After Effects, so I hope a have some sense for timing and working with footage. And I have experience with Virtual Dub Mod and Avisynth. Well, apart from this I'm able to work with Photoshop and Illustrator. Ah, and I'm not a native english speaker (in case you haven't noticed yet)

What would be the right job for the beginning?
And which groups are willing to take newbies in?

Greetings
MB

[Crovax]

Originally Posted by Sylf

I really doubt you need that kind of accuracy, actually. Yes, you may be frustrated with the inaccuracy you see on the wave form. But what counts the most is if the karaoke looks in synch with audio on the screen or not. What I've noticed is that just because I have all syllables timed accurately on the wav form doesn't necessary mean it looks tight and accurate on the screen. Psycho-visual effect, I guess?

What I do is I time it accurately per spoken syllable, and use 'k' tags (with 'kf' it always looks fuzzy...). Then I usually shift all the time indeces forward a little bit, just looks nicer on-screen imho. Prolly serves the purpose of karaoke better too. Obviously need to make the scene-timing adjustments AFTER the shift.


As for Mukashi Banashi...

You shouldn't have a problem finding a group willing to take you on, especially if you are willing to be a timing slave. ;-) 3-4 years of Japanese might or might not be enough to be a useful first-pass translator, guess you're the best judge of that.

Anyhow, if I were you I'd take a look at the 'Help Wanted Classifieds' sticky thread in this part of the forums. See what groups need timers/encoders etc. Or simply find out what groups do a lot of series you like an approach them with an 'open solicitation' or w/e, lol. =P

Good luck!

[kagetsu]

Originally Posted by TheFluff

Uh. No. Tweaking in the +10/-10 millisecond range is pointless. One frame at 24fps lasts for about 42 milliseconds, being any more precise than that (+/-4 100ths of a second) never ever matters since all sensible subtitle renderers needs a new frame to draw a new sub line.


Audio timing on the video sounds rather imprecise to me...

I could care less about the inner workings of how and when its drawn compared to how it calculates its timing. I just did what needed to be done to have it look as good as possible. And most of the time that would be tweaking in that range. Not so much that it was about always looking for "that number" since I just either made it a '0' or '5' when working with video and not audio visualization, the numbers sometimes when you would stop or maybe jump back a quarter or half a second since you needed to stop it exactly when you heard the sound instead of being able to see it cause a little more effort on my part. Some groups I worked for had "rules" to when they would like it timed. Some people preferred half a second or quarter second after the voice started to match the lips usually, some wanted it exactly when the voice was heard. The tweaking was more to just round off to the frame it was going to be showing. I'm a perfectionist and took my work very seriously and sometimes 00:01:56,053 just looked weird, so I would tweak it to 00:01:56,055 and yes that little tweak sometimes did wonders for the timing since going to ',050 or ,060' just still looked a little strange. Like I said, I can care less why, as long as it looked good. Everybody has there own way, and different program usually. And my way worked very well for me, and since I already knew what I was looking for when tweaking it went even faster.

Your right though, video timing is imprecise when timing. But thats why theres tools to help you like going back/forward half a second, a second ect ect. When I did time with audio I still always had to go back and watch my timing with a video display and it was good, but not perfect. So then tweaking that way added time to my work instead of just doing it with video where I got my best results.

To each his own though, but I would never tell someone there doing it wrong since it wasn't my way.

Mukashi Banashi - Basically for me, I was tired of waiting every week for my anime. So I stuck around in the IRC channels, made friends with the ops and when they needed somebody for something I just asked "what does it take to do it?" From there I picked up timing first and helped when the timers weren't around which then lead me to other groups looking for full time timers. I also have a background in English Lit. so I became a timer/ QC'er. Which though still went to a editor before I got the translations. Never got into typesetting though, I was asked to do a series OP once but refused the offer since other people could do very fantastic things with typsetting and I was pretty much on my last leg with fansubbing after doing it everyday for 2 years. Thats pretty much when I got into Raw hunting using Winny very successfully and gaining good contacts with peoples in japan who openly shared their raws with me. So basically, you want to break in. The number one advise I have for you is make good contacts and don't be an a-hole. Because if it wasn't for the people I met and were friendly with. I probably would have never gotten to where I was, which consisted of working with about 12 groups at once timing 3-4 episodes a day. Ohh those days.....

[Sylf]

Originally Posted by kagetsu

I could care less about the inner workings of how and when its drawn compared to how it calculates its timing. I just did what needed to be done to have it look as good as possible. And most of the time that would be tweaking in that range. Not so much that it was about always looking for "that number" since I just either made it a '0' or '5' when working with video and not audio visualization, the numbers sometimes when you would stop or maybe jump back a quarter or half a second since you needed to stop it exactly when you heard the sound instead of being able to see it cause a little more effort on my part. Some groups I worked for had "rules" to when they would like it timed. Some people preferred half a second or quarter second after the voice started to match the lips usually, some wanted it exactly when the voice was heard. The tweaking was more to just round off to the frame it was going to be showing. I'm a perfectionist and took my work very seriously and sometimes 00:01:56,053 just looked weird, so I would tweak it to 00:01:56,055 and yes that little tweak sometimes did wonders for the timing since going to ',050 or ,060' just still looked a little strange. Like I said, I can care less why, as long as it looked good. Everybody has there own way, and different program usually. And my way worked very well for me, and since I already knew what I was looking for when tweaking it went even faster.

Your right though, video timing is imprecise when timing. But thats why theres tools to help you like going back/forward half a second, a second ect ect. When I did time with audio I still always had to go back and watch my timing with a video display and it was good, but not perfect. So then tweaking that way added time to my work instead of just doing it with video where I got my best results.

To each his own though, but I would never tell someone there doing it wrong since it wasn't my way.

Two things.

One - as people has mentioned, the duration of 1 frame at 24fps is roughly 0.042 seconds. When you're making tweaks in magnitude of 0.005 seconds, you're not changing anything as far as the timing of when the sub shows up on the screen. You may think you changed something, but you haven't.

Two - no, nobody times the scripts from wave forms alone. People usually use 2-pass timing method - rough timing/audio timing, where you are mostly concerned how accurately you're placing the sub against when the lines are spoken. This is done much more accurately using the wav form than the video alone. Then they do the scene timing, to better align the beginning/ending of the subs with scene changes etc.

When people talk how wave timing is more accurate, they are talking they can literally see when a character starts speaking, down to centi seconds level, then if they want, they can add some padding. Nobody is denying the use of video in process of timing the script as a whole. People who come from Substation Alpha/Medusa have difficult time comprehending doing accurate timing job with Subtitle Workshop because of the way they've learned to do the initial timing part of the whole job.

Also, hardly anybody uses SRT anymore around here.

[Starks]

Originally Posted by Sylf

Two things.

One - as people has mentioned, the duration of 1 frame at 24fps is roughly 0.042 seconds. When you're making tweaks in magnitude of 0.005 seconds, you're not changing anything as far as the timing of when the sub shows up on the screen. You may think you changed something, but you haven't.

Do not underestimate the power of the human eye. Even still, nobody wants scenebleeding.

[TheFluff]

What part of "nothing will be rendered until the next frame shows up" is hard to understand?

OK, example time since it apparently is hard to understand. Say we have five frames, A-E. To make things simpler let's presume this is PAL (argh) and hence 25 fps, 40 milliseconds per frame. Let's say frame A is the first frame in the video and hence has the timestamp 0. Thus follows (timestamps below each frame in milliseconds):

A  B  C  D   E
0  40 80 120 160

Now, we have a sub line that has the timecodes 00:00:00.035 - 00:00:00.115 (a very short line :V). Depending on the implementation of the subtitle renderer, where this line actually shows up may vary a bit, but let's assume our subtitle renderer won't display a sub line if the start timestamp of the line is later than the timestamp of the current frame (incidentally, this is what VSFilter does).
From this we can deduce that the line will show up only on frames B and C. It will not show up on frame A because at timestamp 0 the line hasn't started yet, and it won't show up on frame D because by then it has already ended. Changing the line start/end times by +/-5 milliseconds may change which frame they will show up on, but it most likely won't. If you want to be sure you actually get any visible change, you should shift at least one frame's worth. In this case, that's 40 milliseconds. Anything less is pointless, because there won't actually be any difference in how it's rendered at all unless you're lucky and hit a frame timestamp change.
Actually if you're doing this kind of thing at all, you're much much better off just timing on the video frames directly because then you're sure about how it will look.


As for the karaoke timing thing, it's the same. Noone will ever notice any difference, especially not if the syllables are shorter than the frame length. I consider my eyes to be rather more accurate than the average viewer's, I can catch one-frame combing artefacts in full 24fps motion, and I can safely say that plus minus one frame when karaoke timing is well inside the "noone would ever notice much less care" area. Human judgment on audio sync is definitely not all that accurate...


Edit: fun fact: even if SRT actually uses millisecond precision, rendering it with VSFilter won't actually give you millisecond precision anyway since VSFilter converts everything to (centisecond-precision) ASS internally.

[False Dawn]

Originally Posted by Sylf

I really doubt you need that kind of accuracy, actually. Yes, you may be frustrated with the inaccuracy you see on the wave form. But what counts the most is if the karaoke looks in synch with audio on the screen or not. What I've noticed is that just because I have all syllables timed accurately on the wav form doesn't necessary mean it looks tight and accurate on the screen. Psycho-visual effect, I guess?

That's actually quite an interesting point. Even if you time the syllables/words exactly the when the vocal starts, it will always seemingly appear late onscreen. Having noticed that it doesn't really do the same when it flashes off, I can only suppose that there's some kind of mismatch in the sub-timing/onscreen-image - which is why a number of timers use lead-in to counteract that. Or perhaps, it is something to do with the eye not being quick enough. Who knows?

[Starks]

Originally Posted by False Dawn

That's actually quite an interesting point. Even if you time the syllables/words exactly the when the vocal starts, it will always seemingly appear late onscreen. Having noticed that it doesn't really do the same when it flashes off, I can only suppose that there's some kind of mismatch in the sub-timing/onscreen-image - which is why a number of timers use lead-in to counteract that. Or perhaps, it is something to do with the eye not being quick enough. Who knows?

Everyone has different tolerances for accuracy. Mine's 0.25 seconds.

[Quarkboy]

Hehe, Thefluff seems to have forgotten about how vsfilter renders \t effects...

although there would only be a frame rendered every .04 seconds or so, when you do the mathematical interpolation you actually need to keep a half or even quarter frame accuracy to avoid subtle rounding errors.

In truth, it's the same principal as q-pel.

[TheFluff]

Originally Posted by Quarkboy

Hehe, Thefluff seems to have forgotten about how vsfilter renders \t effects...

although there would only be a frame rendered every .04 seconds or so, when you do the mathematical interpolation you actually need to keep a half or even quarter frame accuracy to avoid subtle rounding errors.

In truth, it's the same principal as q-pel.

Obviously effects are a completely different bag of tricks; I was talking about plain \k's. If you're doing effects it gets a lot more interesting, but still, if the entire syllable is shorter than about 1.5 frames or so and the actual effect doesn't cover any more than that time, it still won't matter much since you'll only see a single snapshot of it. Of course sometimes you have effects that cover much more time than just the syllable itself, then different rules apply.

[Starks]

Originally Posted by TheFluff

Obviously effects are a completely different bag of tricks; I was talking about plain \k's. If you're doing effects it gets a lot more interesting, but still, if the entire syllable is shorter than about 1.5 frames or so and the actual effect doesn't cover any more than that time, it still won't matter much since you'll only see a single snapshot of it. Of course sometimes you have effects that cover much more time than just the syllable itself, then different rules apply.

Ideally, yes. Realistically, not really.

You're just splitting hairs now.

[TheFluff]

Originally Posted by Starks

Ideally, yes. Realistically, not really.

You're just splitting hairs now.

what

Edit: Seriously, I have no idea what the heck you mean with that.

[dj_tjerk]

Originally Posted by False Dawn

That's actually quite an interesting point. Even if you time the syllables/words exactly the when the vocal starts, it will always seemingly appear late onscreen. Having noticed that it doesn't really do the same when it flashes off, I can only suppose that there's some kind of mismatch in the sub-timing/onscreen-image - which is why a number of timers use lead-in to counteract that. Or perhaps, it is something to do with the eye not being quick enough. Who knows?

It is something with the eye being slower to notice change than your hearing is.. [reference needed]. So you need to shift everything 0.05/0.06s earlier to sync your eyes with your ears.

[chaos4ever]

Originally Posted by dj_tjerk

It is something with the eye being slower to notice change than your hearing is.. [reference needed]. So you need to shift everything 0.05/0.06s earlier to sync your eyes with your ears.

Sight integration is about 33Hz. I wouldn't be surprised if ears integrated signals much faster than that, since ears can resolve stereo differentials on the order of about 1ms, which would give an upper speed limit of about 1kHz.

[cyth]

Originally Posted by chaos4ever

Sight integration is about 33Hz. I wouldn't be surprised if ears integrated signals much faster than that, since ears can resolve stereo differentials on the order of about 1ms, which would give an upper speed limit of about 1kHz.

What about sound traveling slower across space than light?

[chaos4ever]

Originally Posted by Toua

What about sound traveling slower across space than light?

sound travels about 300m/s, so we'd be looking at about a 10us to 10ms delay before it hits the eardrum, depending on how far away you are from your sound source. sight integration is much slower than that, despite almost instantaneous signal propagation to the eye.

btw, further thinking makes me believe that, sound integration may actually be dependent on frequency of sound among other things, so the 1khz may not be a valid cutoff, but just a rough approximation on my part.

[Talon]

Hello everyone, I did a search of the forums and came up empty on this so I thought I would start a thread.
I have been into anime for a very long time and I recently decided that I wanted to help out with fan subs but the problem is I have no idea where to start! I don't even know all of the different jobs required to be filled to create a fan sub. I have no real back ground in this area but it has held an interest for me for quite a while now and since I have a LOT of free time on my hands I thought this would be a good project to get into.
In my defense I am a quick learner and get almost obsessive when there is a new skill for me to learn. any advice would be greatly appreciated, I really do want to help out after all, if there are any groups out there looking to recruit and train someone like me I would be most receptive to the offer.

I guess thats it for now, thanks for your time everyone.