ep1

>>149
yes, although since ass is plain text this means that effectively any plain text can be attached as long as you name the file properly. Not sure if the same holds for other files since server might verify the file format magic number (e.g. GIF87a for gif)

Attached is ep2. For this one I found rebasing the other way was cleaner, cherrypicking BB subs onto CR release, since the timing/flow of the CR release was better.

Oh and if anyone wants to practice JP, seems like japanese netflix has bocchi the rock with subs, but you can also find episode transcripts linked with key frames here (and also read reactions from jp netizens)

ep1 http://anicobin.ldblog.jp/archives/59802332.html
ep2 http://anicobin.ldblog.jp/archives/59825029.html

Nevermind looks like someone else has already committed themselves to merging CR/b-global https://nyaa.si/view/1591618, so I can stop doing this. Diffing ep1 between the one here and theirs they incorporated the bulk of the cases where b-global was better, although I feel they didn't fix the overlocalization enough and didn't split multilines where I would have personally preferred. E.g. by line-slitting I mean

Gotoh sub:

>Could someone like me \N shine like them if I started a band?

Sub attached here:

>If I started a band, perhaps even someone like me
>Could have a chance to shine!

JP
>bando kundara… moshika shitara watashi mitai na ningen de mo kagayakeru

where the sub here mirrors the JP sentence structure so I feel it's closer to 1:1 and makes following along easier. There's several cases such as this where one of the existing BB/CR sub can be trivially reworded to match JP clause order.


And in terms of localization

Gotoh Sub:
(unchanged from b-global)
>But before the neighbours report \N my shitty song as late night noise pollution

(unchanged from CR)
>I look... super cool!
>I'm already a total rocker chick!
>I look like a real somebody!
>The people can't look away!
>Did my midnight high make me overdo it?
>But I'm putting out such \N powerful rocker chick vibes...

(Somehow made even worse than the original sub, as for some reason the author commits the unforgivable sin of changing "cringy" (which overused a word as it is is at least a proper adjective) into "cringe").
>Right!? That name's kinda cringe though.

(Somehow made even more localized than the source subs)
>every big guitarist was practicing \N their butt off behind the scenes,
>even when there wasn't \N an audience in sight for them.

(Unchanged from CR)
>I'm gonna get over my social anxiety
>and bring out my full guitarhero powers!

Sub attached here:

>But before I cause late-night disturbances \N playing out my shitty original song

>N-neat...!
>All of a sudden I've got the band-girl look
>Like a real somebody!
>My presence is amazing!

(I regret to use the word cringy here but I'm drawing blanks on a suitable synonym that's as succinct)
>Though her naming sense is a bit cringy...

>is that even awesome, popular people
>must put in a lot of work where we can't see.

>I'm gonna get over my social anxiety
>and show them what I can do as Guitar Hero

>>154
Nevermind, somehow the "gotch" guy did identify the lines where b-global was better but decided to rewrite half of them in so-called "zoomer" slang. The release group should have been named "[botch]" instead. I mean look at this

>You should practice what you preach.
>Put the popcorn away.
>I can't sing for squat
>The heck is this about
>Less goofing, more working.
>Your face has suspicion written all over it, y'know.

These are somehow worse than both CR and b-global. I guess I'll have to do the subs myself after all.

>>/an/36610
>>153
Thanks for linking that site. It's a useful resource.
>>148
>this is the only board which allows text file attachments.
What do you mean? There's also /ddl/, /ma/, /navi/, /tat/, /test/.
But /sub/ is fine too and makes the most sense along with /an/ and /ddl/. Just keep in mind that the files you upload could get overwritten, assuming this bug hasn't been fixed yet: >>/ot/36524

>>162
Ah you're right, I never noticed that /navi/ (amongst the others you mentioned) allow text files. But yes as you mentioned this board makes the most sense, and I don't want to clog the main boards. It's a bit less discoverable, but so long as there's a link to it from /an/ people should be able to find it. (Although I realize I forgot to link this in the dedicated bocchi thread that was created; I'll probably do that once more subs get added.)

>Just keep in mind that the files you upload could get overwritten
Thanks for reminding me, going forward I'll try to prefix episode number and show title to avoid collisions. I do also plan to upload these to IA when done so they'll at least be preserved somewhere and indexable.

I'm pretty surprised by the typesetting effort on ep3. At first I thought it was the "gotch" fellow's doing, but no it's present even in the original CR release! It's not every day that you see corporate subs use rotation, skew, color, AND fade all at the same time! Although to give credit, the "gotch" guy did fine-tune fade, color, and font to make things blend in a bit more, which is more than I could ever do.

Post edited on 31st Oct 2022, 3:35pm

Epsiode 3: edited version of the "gotch" subs. Honestly with how much editing was done I'm not really sure using it as a base over CR really provided any benefit. The major changes were undoing retarded overlocalizations, and adding back in honorifics where it fits.

In particular for honorifics, since the gotch version had them stripped out entirely I opted to only add them in lines that were short enough to accommodate them. I.e. a hybrid approach where the honorifics are elided in full-sentences but preserved for interjections. Also for some reason the "gotch" guy also stripped out "senpai" everywhere even when it wasn't used in an honorific sense.

It feels like this one took longer to do than the uchi no shishou subs because I kept flipping back through the 3-way diff. Going forward the general pattern will probably be to assume that the "gotch" subs have had the relevant b-global lines folded in (which seems to be generally true), and diff with CR only, correcting honorifics and undoing retardization along the way. That should allow for a one-pass editing approach.

Also there's a golden opportunity for a subgroup to call themselves "unbocched" (get it?). The last time you got a pun opportunity like that was for stella no mahou subs (daikirai or die-suki, in response to the infamous daisuki subs).

Ah finally I found some delicious sub drama with people talking about the gotch release (to be clear, none of the people quoted below are me). [And note that I only wish to record the below not as kindling for a flamewar but rather to get a pulse on the zeitgeist of the anime community with regard to subs]

Anon 1
>I download it earlier to to evaluate the translation and the result is disappointed.
>Despite the subtitle has good animation, the english translation is bad. they even remove "-chan".
>they also make 1 mistake where Nijika supposed to said Hitori instead of Botcchi.
>how "Please return back to Earth Hitori!" became "Earth to Botcchi, please!"
>Like how "Let's calm down" ended up becoming "Let's not be to hasty!"
>the worse part when Nijika: "prickly, spiny, spiky, thorny but really nice" instead of "tsun-tsun-tsun-tsun-tsun-tsun-tsun......dere.
>There's high chance this translation was created to suit with western view.

Anon 2
>First and foremost, what's your japanese proficiency level? Surely N12.
>The TL is a merge between CR and B-Global scripts, and apparently both suck according to the description on nyaa, only thing Gotch did was remove honorifics

Anon 1
>I can agree that Gotch has better encode but that translation still not my taste.
>and of course I notice that Gotch use CR/B-global script then reedit to suit him but that's okay to me.


Anon 3

>Are you stupid? Seriously. I know honorifics are a preference (I personally don't care for them, but I don't mind if subs have them, but using "Please return back to Earth Hitori!" to "Earth to Botcchi, please!" as an example of bad translation is so fucking stupid. Maybe English is your second language, but those both mean basically the same thing. If I say "こんにちは元気ですか" you don't know what that means, but if a translation said "Hello, how's your day been?" for it, you'd assume that was fully accurate, but "こんにちは元気ですか" could also be "Hello, how are you?". And you might think that those basically mean the same thing, which is true, but "Please return back to Earth Hitori!" and "Earth to Botcchi, please!" are the same thing, since they both are based on the same kanji, but use different words to convey them to the viewer. That is to say, a differing translation isn't a bad thing, it's just a different representation of the kanji being spoken, and as long as you get the meaning behind the sentence, no matter what words it uses within that sentence, then the translation has done its job. (This same argument can be used for "Let's calm down" becoming "Let's not be to hasty!") Also, ASW and Yuisub are not good, as well as amZero, no matter what you say on the translation, either Gotch or SubsPlease are the best encodes for this anime, and to say otherwise would just be objectively wrong.

Anon 1

>But for translation, I like it more when translation is more accurately translated based on what JP voice dialog actually said it.
"Earth to Botcchi" meaning isn't wrong. Besides, this isn't the first anime using it. But I can't feel the translated connected with the way anime character talk. I mean every anime character has their own personality dialect and so on.......
Like example a character that always end her dialog with "-desu" (Death)
the translation will certainly put off the view if the word -desu or death didn't even appear on translation despite it will not messed up the meaning.
>btw, the reason "Let's not be too hasty" isn't the only wrong because the manager replied it with wrong as well "You should practice what you preach" instead of "You're the one that shouldn't be too hasty" or "You're the one that need to calm down"

>PS: I'm not watch anime for just a single or 2 year but more than 10 year. Even I can understand what JP voice want to express so the conversation connected with each other despite not learning JP language properly. That's why I think there's high chance that Gotch translation was edited to suit with western view. Meaning that Gotch way of translation may for certain western country. Btw, I'm from Eastern Asia country.

Anon 4
>English translation is made primarily for those whose first language is English and those who are about as fluent in it as native speakers. You clearly aren't. In general, what you consider "accurate translation" either loses subtextual meaning to being overly literal, or makes for awkward phrasing and janky, unrealistic dialogue flow. Dubs are even more strict on flow and phrasing and will take further liberties to ensure every line sounds like something a native speaker could potentially say in that situation.

>The problem with your argument is that, just like anyone who has never been involved in professional translation, you get overly preoccupied with translating *words* as opposed to translating meaning, intent, and tone, all of which are arguably better captured in Gotch's edit than your suggested changes. Besides, it's always funny to see translation accuracy judged by people who aren't good enough at either English or Japanese to have a qualified opinion. If you want a fully accurate experience, just learn Japanese and turn off subtitles.

>>171
(responding to myself)
I find it interesting how many people chime in without bothering to diff things themselves. E.g. the claim
>only thing Gotch did was remove honorifics
Which is not true, lines were rewritten (i.e. gotch release has lines that are in neither source)

Also find it fascinating that people care so much about the encode as a reason to pick one release over another. Do people not know how to extract sub tracks? And that entire long-winded rant about how "earth to bocchi" is the same as "please return to earth" is hilarious consdering the original dialogue is "現実逃避しないで" and to me at least there's no need to use "earth" at all since "stay here in reality" sounds fine to me (though if you're uber pedantic, the original JP is a negation).

I think this might be my new favorite phrase of infamy
>as long as you get the meaning behind the sentence, no matter what words it uses within that sentence, then the translation has done its job

And as for this:
>either Gotch or SubsPlease are the best encodes for this anime, and to say otherwise would just be objectively wrong
First the Gotch fellow didn't encode it seems to have been done by a different guy (no documented association between them, but they might be in cahoots who knows). In terms of the actual image-processing that was done, it's apparently

>Light denoise, a bit of debanding, stronger than usual dehalo and medium ish AA.
I don't have enough experience on practicalities of being an encoder so I can't comment, but again I find it interesting how people are so fixated on video quality here. Client-side shaders are good enough that they can be run in real-time on top of the source, so unless you're freeze-framing and A/B testing I'm skeptical that you'd see any actual difference at standard viewing distance. Also given that post-processing can always be done from the source (but the post-processing itself is a one-way operation) for archival at least it would seem that storing the source is better (especially when the source for these encodes is I assume the webrip itself). I'll give the benefit here and assume they're referring to "objectively better" in the sense of perceptual quality for a given filesize in which case yes I'd agree that a re-encode of a webrip can probably be much smaller than the source without sacrificing quality via tricks such as debanding, client-size dithering (10-bit), anti-aliasing, etc.

Also I've only come across anti-aliasing in a signal-processing context where you low-pass prior to sampling (or equivalently in CG where you supersample then average). I'll need to read more about anti-aliasing in this context where you've already got the sampled image. I assume it'll somehow end up being related to doing blurring or interpolation on the existing image.

Anyone got suggestions for translating
>senpai no musume
which Kita says? (see http://anicobin.ldblog.jp/archives/59846597.html for context if you need a refresher on the episode).

I'm assuming (which could be wrong, this is a construction I have not really heard before so I don't know if it's common) that this plays off of a dual-meaning of musume both as a generic polite word for young girls in general, as well as the meaning of daughter; so "senpai no musume" might be equivalent to "right-hand girl for senpai" but also with the implied meaning of a yuritease kin-esque relationship. Translating just as "senpai's daughter" doesn't work because in English "daughter" seems much more heavily tied to the family relationship and doesn't stand by itself (i.e. if you see a random person on the street, you can only call them a daughter with respect to their parents). "Senpai's girl{|gal?}" does mostly capture the two meanings but sounds strange to my ears, a bit slangy. "Senpai's fille" might work if everyone knew french/latein.

Then again maybe I'm overthinking and "senpai no musume" is a strange construction in JP as well.

>>173
I think it's only being used in the "daughter" sense here. Bocchi reacts the way she does because it is somewhat strange to want to be the daughter of the senpai you have a crush on.

These Japanese Twitter posts clearly construe "娘" as "daughter" in this line:
From your link:
>山田は私の母になってくれるかもしれない女性だ！
https://nitter.net/Diamond_Diagram/search?f=tweets&q=%E5%A8%98&since=2022-10-15&until=2022-10-16&near=
>後藤、完全に山田と伊地知の娘だ……
>喜多ちゃんも「リョウ先輩の娘になりたかったの……」って言ってたし、親子型のダブルペアでキャラ括ってるのかこれ

More relevant discussion can be found in these threads:
https://desuarchive.org/a/thread/244135440/#244169876
https://desuarchive.org/a/thread/244193424/#244197631

>>172
Ok I think I roughly get how post-sampling AA works. Basically recall that when you sample a non-bandlimited signal, the frequencies above nyquist get folded/wrapped down and serve as high-frequency artifacts. Now like any signals we can do fourier-domain filtering on this to try to remove out the high-frequency artifacts and recover an approximation of a band-limited signal. Clearly we can't do this exactly since we don't know what the original frequencies above nyquist were and so can't construct the appropriate low-pass filter, but we can probably make some assumptions and get close. This is how FXAA seems to work, it seems to do the equivalent of choosing an appropriate threshold by analyzing the image, and then low-pass filtering (of course this is done in spatial domain not frequency domain so it's a simple convolution).

Now for the case of encoders in particular, who knows exactly what the mystery scripts do, but at least one group was kind enough to actually document their filter chain, but "insaneAA" seems quite overkill. Maybe can analyze something simpler. Most antialiasing scripts for Avisynth utlimately seem to call into SangNom or nnedi3.

For example, "naa()" script: does

Clp.nnedi3_rpow2(rfactor=SS,cshift="spline36resize",qual=2) .nnedi3(nns=2) .TurnRight().nnedi3(nns=2) .TurnLeft().Spline36Resize(OX,OY)

which seems to be nnedi3 upscaling, following by intra-frame interpolation on both x-y, then downscaling.


>SangNom is a single field deinterlacer, no fancy stuff, just plain edge-directed interpolation.
>nnedi2 is an intra-field only deinterlacer. It takes in a frame, throws away one field, and then interpolates the missing pixels using only information from the kept field

I was not familiar with edge-directed interpolation, see reference for more. I can sort of understand why this works, interpolation on low-frequency content will be unchanged, while interpolation on high-frequency content will smooth it out. Feels very heavy-handed to me though, if you have genuine high-frequency content seems like that would be affected as well since the filter has no spatial localization. I don't know much about dsp or filter design, there are probably better ways to do this.

https://developer.nvidia.com/gpugems/gpugems2/part-iii-high-quality-rendering/chapter-22-fast-prefiltered-lines
http://www.cse.chalmers.se/edu/course/TDA362/aliasing.pdf
https://www.wepc.com/tips/anti-aliasing/
https://github.com/Beatrice-Raws/VapourSynth-insaneAA
https://chiranjivi.tripod.com/EDITut.html

>>174
Got it, thank you for the links.

Episode 4. Uses CR as base, both since gotch is not released yet and because I see no benefit script-wise in doing so (typesetting is a different issue, I might backport that if it's significantly better than stock CR).

Overall I think CR script for this was mostly fine, the majority of edits were stylistic preferences, slight delocalization, and single-word substitutions. I would not say that this is an order-of-magnitude improvement over the original (unlike for ep1 and ep2 where I do feel that the edited version is markedly better than CR source). I probably spent too much time bikeshedding English nuances and etymologies.

Episode 5.

Episode 6 attached. Also I found b-global's translation of one of the lines as
>Did you blue screen out of joy?
absolutely hilarious given the visual gags in the previous episodes (but I don't think she meant that usage of freeze since she never saw bocchi undergo datamosh). I really like the new character introduced, seems like Bocchi is able to relax and open up around her.

>>175
More on post-sample AA (which I prefer to think of as post-processing shaders or filters). It was still unclear to me how exactly this looks frequency-domain wise, i.e. how it differs from a low-pass filter. I think the key is that unlike a traditional gaussian filter where you use a fixed kernel, morphological AA methods use an adaptive kernel depending on edge constrast and edge direction. I.e. to deal with AA by blurring, you need much stronger blurring in direction perpendicular to the edge than you do in the parallel direction.

Here's some relevant links for people who'd like to learn more: I think the original intel paper on MAA is a good place to start.

https://web.archive.org/web/20130213010835/http://visual-computing.intel-research.net/publications/papers/2009/mlaa/mlaa.pdf
https://www.realtimerendering.com/blog/morphological-antialiasing/
https://blog.codinghorror.com/fast-approximate-anti-aliasing-fxaa/
https://www.overclockersclub.com/reviews/serious_statistics_aliasing/9.htm
https://catlikecoding.com/unity/tutorials/advanced-rendering/fxaa/
https://developer.download.nvidia.com/assets/gamedev/files/sdk/11/FXAA_WhitePaper.pdf

>>180
I don't know if anyone's actually using or waiting for these, but just a heads up that I don't think I'll continue making them for ep7 onward. It's popular enough that some group will probably do it for the BD release, and the edits for the past few episodes have been mostly subjective style corrections.

Just wanted to say thanks for doing this. I've tried to use these subs with every ep I've watched so far (Though I'm not entirely sure if it's properly switching from the built in subs)

>>185
Thanks, it's good to know that someone at least found them useful. And for me it was good practice to set up a workflow so I could really put effort into the Uchi no Shishou subs.


> I'm not entirely sure if it's properly switching from the built in subs
Oh you should be able to tell that pretty easily by just comparing the known dialogue. E.g. for Ep1 at 0:05:13 my subs used "But before I cause late-night disturbances\Nplaying out my shitty original song" whereas the original CR was "But before the neighbours report \N my shitty song as late night noise pollution.


Also amusingly seems like the Gotch fellow stopped making subs after that 3rd episode. How peculiar. Either way, this has convinced me nyaa has done more harm than good in effectively "centralizing" anime distribution and then leaving it to rot (i.e. disabling account sign ups).