It's been five months, so maybe a little housekeeping is in order?

Nah.

On occasion, I will step through certain scenes in an episode (or OVA) frame by frame. It's not unlike firing up a debugger, and I could see the troubleshooter's attitude being applied here, but I'm merely a curious onlooker. If I wanted to play QA, I would join a fansubbing group.

A leisurely stroll through the most kinetic scenes of Kara no Kyoukai probably operates at the same level as those slow motion highlight reels of … anything, really. This highlights the sobering fact that my internal refresh rate (or shutter speed) is a lot lower than 23.98 fps, but by underclocking the input I can truly gawk at the whole spectacle, and maybe even grok just a bit of it.

Paradox Spiral is the goto movie for technical achievement, which is a euphemism for "details." I don't think you have to play the "gotta catch them all" game to know what's going on — capturing 63.2% on the first pass is more than sufficient — the reason being that what you missed is likely covered by what you did see by the time the credits roll.

That there is a lot of subsequent hand-holding doesn't make finding these things any less fun. Case in point, the scene where "Shiki" (in quotes for a reason!) escapes from her magical prison is foreshadowed in a few lines spoken by Touko, is insignificant but for the very last shot, but is still kind of neat.

This space shuttle frame had enough contrast that I could spot it in real time, but if you step back to the very beginning of the sequence …

Like I said, the fact that Shiki is perusing the mother of all CCTV recordings was given away previously, but dinosaurs are cool whether they are anticipated or not. Wait, did I say dinosaurs?

There we go.

An anonymous 20th century battlefield. It's a bit of a timeskip.

Not the most painful image through five films. Seven, even.

There is more, lots more, but in the flashback vein. Except this one:

Where is this from? Who is portrayed? Does it matter? I'm thinking no, it doesn't, but it would be nice to know regardless.

Push-button encoding at its finest, especially now that x264 has presets.

This is the second and most current televised performance by Kalafina. I may eventually get to the first one, but if I never do, it is sufficient to know that Kalafina @ Anipara Ongakukan was unmitigated disaster. Except Hironobu Kageyama. Maybe.

Before going any further, it's worth noting that some of the performers had in-ear monitors, and Kalafina wasn't one of them. For those without IEM's, usually there are floor monitors at the front of the stage, but the IEM-less happened to be monitor-less in general, except JAM Project.

It's not an excuse, but not being able to hear your co-singers well, let alone yourself, is a definite handicap. For what it's worth, I found that the IEM-equipped singers — ELISA, Yousei Teikoku, Yui Horie, and Nana Mizuki — had the most technically proficient performances. Nana Mizuki also had two floor monitors, and Yui Horie had one IEM per ear.

Kalafina had two songs on the show, in theory, but ARIA was cut. Unless ARIA was a poor showing, I'd have run ARIA in a heartbeat over 光の旋律 (Hikari no Senritsu), despite the latter being an OP for the currently airing Sora no Oto.

Hikari no Senritsu is a much lighter song, with sparse accompaniment and a thin texture that isn't any easier to pull off. Supposing they do, the impression they make is minimal. And it's a given that they don't know Hikari no Senritsu nearly as well as they know ARIA. If Kalafina wants to promote, ARIA is the better gambit.

Anyway, this clip starts with an introduction/greeting statement and I don't know about you, but it appears to me that the expression on Hikaru's face is some combination of exhaustion and nerves. This is precisely the condition that you want to be in beforehand.

Whether or not any of them are actually tense or tired, tentativeness reigns everywhere except for the two choruses: Hikaru's first verse is a bit tight and it's difficult to find much articulation in her phrasing; the first bridge is littered with early and late entries as Keiko and Wakana are never quite dialed in. Balance and intonation are just marginally convincing because Keiko is trading pitch control for a softer dynamic level. I can't think of any situation where this would be a good exchange.

Things get moderately unhinged in the second round, when Hikaru flubs the entry and fades the last note of the verse. Second bridge is much more locked in but is marred by a string of bad intervals, the most obvious example coming at 2:27-2:29 where the music just seems to wilt.

ANISONG SP2 and other poor outings have caused some to question Kalafina's singing ability, while getting angry about it (i.e. kacpy). I'm ambivalent about the whole thing, because it's par for the music show course, and I don't think that being ill-prepared and ill-equipped should result in the application of some upper bound on what people are capable of.

Besides, it could have been worse. Like, Lion.

P.S.

Okay, I concede that May'n was more or less pitch perfect on pink monsoon and Megumi Nakajima was causing all the problems on Lion, so maybe I should say the above list is a list of proficient performances that I liked.

I forgot that 1440x1080 is actually 1440x1088 and wound up not cropping the bottom before resizing. Oops.

I bet this has been done before, too.

People like to talk about how good Baccano! is. Or should I say, was? At any rate, I don't get the impression that there were many boosters for the show even during its TV run, and what praise they did heap on was pedestrian in nature.

Two common selling points I've heard for Baccano! are that it is well written, and there's a lot of stuff going on. Combined, you might say that the show is technically solid, but that's the same thing as not screwing up the juggling act.

Suppose you made it through the day without tripping over your shoelaces. Should you receive a medal? I thought we had higher standards than that.

The act of time multiplexing several different stories is itself a structural flaw. It's expected that gears get changed eventually, but eventually is not the same as frequently. Run up and down the gearbox quickly enough and you enter the realm of multitasking.

You are free to think that holding ten different conversations across MSN, IRC and GTalk makes you a great multitasker, but I'm going to have to defer to Stanford, whose researchers claim that humans are generally terrible at it.

The kind of non-linear storytelling Baccano! does, if converted to text, would require that every fourth paragraph begin with "Meanwhile, in 1930/1/2," and would invite totally justified accusations of being difficult for the sake of being difficult.

Schizophrenic interleaving also has this nasty habit of killing expectations. The deal with mind and multitasking is not that you miss stuff. On the contrary, everything registers. Your ability to focus, which is a prerequisite for your ability to care, is a different matter.

I suppose that means I did not care for the characters too. The crazy ones did stand out as a matter of course, including (unfortunately) the crazy stupid ones. But those we ought to be sympathizing with have their tragedies read off like a bullet point list. Told sequentially, we would call it a Key arc, but jumbled up somehow made the show one of the best of 2007.

At some point, all that stuff getting filed away each episode becomes relevant, but you'll never have to think too hard to figure it out. Baccano! just happens to be obfuscated spoon-feeding, and those moments of realization are invariably accompanied by the show telling you to say "Ah."

I appreciate the sentiment of No Viewer Left Behind, even after the show tries to, uh, leave the viewer behind.

Cute attempts to justify the show's structure should instantly raise red flags. This is a show, making an elevator pitch, before it even gets underway. If the staff is confident in the layout of the story, the show should be the sales pitch, not have a sales pitch.

Sorry, but there's no blood to be drawn from this stone. Not without a whole lot more signal.

P.S.

The fact that the anime is based on a light novel series is not lost on me.

Oriental Riff sighting: Episode 8, second half. Still can't believe they have a name for it (hat tip).

I am sure that something like this has been done before, elsewhere, but I can't be bothered to look.

The Demoman

The Spy

The Heavy Weapons Guy

The Soldier

The Pyro

The Sniper

The Medic

The Scout(s)

The Engineer

Appendix

What on earth am I on about?

There are/were some who are/were dissatisfied with Clannad (2007) because ~After Story~'s reset ending trivializes the events that led up to it. Logistical — and logical — issues aside, so what if those events never really happened? Why should you feel differently one way or the other?

Way back in the day, there was a kid who had a jerk of a dad not unlike Naoyuki Okazaki. Incidentally, this kid's father also wound up in a jail of sorts: debtor's prison.

…

*cough*

…

So anyway, as I was saying, Ebenezer Scrooge saw a vision of the future that — wait for it — never really happened, seeing as how it was a vision and all. Now our old chap Charles did have a tendency to be generous with the verbiage, but my word! What a waste of words!

Scrooge should have lived through Tiny Tim's death, died of grief himself, had his ghost bear witness to all the acts of ill will towards him, and then begged his maker for a second chance. Only then would the story have any merit. Only then could it be called a classic. Right?

I think what people despair about is the perception that certain kinds of endings steal or otherwise undermine their emotional investment. It's true in cases where the story ends at the start, but what about this one? If you think so, I know where you can go to retrieve your investment; it's in the vicinity of where you voluntarily threw it away.

Abstractions and illusions may not be the most tangible things in life, but anyone familiar with imagination and language must acknowledge their transformative influence. Whether or not you believe Tomoya's trauma and tribulations were real, the ordeal changed him regardless. Rather than disparage it, let us salute Tomoya's true graduation into society.

A fanfare, good conductor, for the Common Man.

P.S.

TV Tropes always has some insight. I think it's worth checking out:

• Bait and Switch Credits
• Identity Amnesia (makes the claim that Nagisa tests Tomoya before coming back)
• Story Arc (one-arc stands)

The first thing I did after finishing Clannad ~After Story~ was to go searching for that timeline diagram that I forgot to bookmark way, way back.

Was it necessary? Not at all, but it's nice to have some external information in order to frame the ending as something resembling coherency.

Certainly it's critical that you give yourself to the emotional roller coaster in order to ride it, but now that the ride's over, I want off lest I be subjected to 4+ g's over and over. Comprehension is closure, in a sense. And when there's nothing there, I'm content to know at least that much.

I have never played a visual novel, but I've read more Choose Your Own Adventure books than I care to admit. As silly as they were, they would have been even sillier had the various paths been linearized, and not nearly as interesting to nine year olds.

There are some things that can be done, but short of inserting numerous resets, the end result is going to involve a steamroller. This is most obvious, at times painfully so, in Air (2005) and Kanon (2006), neither of which got far from the feeling that they came about via a 20 ton mass.

Clannad (2007), too, if you view the first season in isolation. But ~After Story~ wouldn't be ~After Story~ without the Clannad moniker preceding it. Where Clannad as a whole succeeds is in the integration of its cast. Characters pop up just often enough to do or say something meaningful, showing us — and each other — that they are more than token acquaintances.

It's because the source is about depending on the kindness of friends and family that it steamrolls unrolls better than its predecessors, which is to say, less one-arc stands. But what separates it from most other high school rom-sitcoms is that the "Where are they now?" aspect is not relegated to the finale, as some fanciful backdrop for credits and an extended ED.

So when it comes to time frame, scope and premise, you could do worse than a game of life.

Appendix

• The title screens acted as a loop counter, provided you were paying attention. I wasn't.
• In ~After Story~ Episode 16, Nagisa passes Tomoya on his left. In Episode 22, she passes on his right. Significant?
• Fuuko's role notwithstanding, the diagram is a lot easier to accept than this theory, which is good because that place is difficult to navigate, let alone read.
• I guess this entry sort of mirrors this one.

Some fine folks over at CPM forums have subtitled an interview that Yuki Kajiura gave in 2003. I once saw the original floating around on YouTube, skipped around a bit of the video, but never made it to the end because of this language barrier thing.

The interview is the most personal interview that she's given, at least of those that have been translated. It's probably the only time you or I will ever see her smoking, for one. Family photos abound, and she's candid discussing the death of her father.

What is most fascinating is the end of the interview that I never got to, a piano rendition of Akatsuki no Kuruma that preceded — or foreshadowed? — the formation of FictionJunction YUUKA. This interview perhaps inadverdently captured the entirety of where she came from and where she was going at that very moment. I just found that really neat.

And it just goes to show you, even an English Literature major can be an engineer (in the unlicensed sense of the word).

P.S.

Secret to success: rope your child into being your own karaoke machine.

Appendix: Some (translated) interviews

1. From 2002
2. From 2008. Low points coincide with her father's passing
3. From 2009. Artsies rule the world?

Every pipe needs a source and sink. Or is that the other way around?

Consumers

I didn't say anything about it at the time, but I was one of those people who took issue with HD fansubs in general, and the H.264 codec in particular. Then, my laptop was just over 5 years old. It was pokey by 2007 standards, but in 2002 it was almost a high-end desktop replacement, sporting a 1.6 GHz Pentium 4-M processor.

That made the CPU NetBurst based, and it ran about as hot. So hot, in fact, that the P4-M conspired with poor component layout to wreck my hard drive just before the one year warranty expired. I learned my lesson and later installed external fan control, plus kept the CPU at 1.2 GHz for most tasks, including playback of SD fansubs based on XviD/DivX.

Codecs based on the MPEG-4 Advanced Simple Profile (ASP) were codecs for a simpler time; the fact that you can get a DVD player that does DivX playback for less than $50 is proof of that. Meanwhile, I was quite happy with hardsubs in XviD in an AVI container because that meant I could keep the laptop's two fans from screaming at me.

Fansubbers internally had been unhappy with this arrangement for quite some time. They disliked the AVI container because it made handling variable framerate video difficult, something that usually arises due to the OP and ED being animated at framerates different from the episode itself. On top of that, AVI did not play well with soft subtitles, so any revisions meant another encoding stage.

There were also rumblings about ASP itself. Compression is fundamentally a trade-off between processing time and occupied space, and the designers opted for quick processing time. Storage requirements increased as a result, but at SD resolutions this never got of hand. H.264/MPEG-4 Advanced Video Coding (AVC) never gained traction until consensus deemed AVC better suited for HD relative to ASP, and then people suddenly realized that H.264 worked for SD subs as well.

I suppose Arienai's subs of Hataraki Man should have been a shot across the bow: 640x360 H.264 in an mp4 container resulted in 80% CPU utilization, and the thing was still hard-subbed. Loved the series, but I did have to turn the volume up. Adding MKV and soft-subs made SD releases nigh unwatchable, at least not without opening the throttle up to 1.6 GHz. In high motion scenes, there was no avoiding video stutter.

HD content also proved unplayable, but for a different reason. I had transport streams and DivX AVI's of singers (read: Utada Hikaru) doing the promotional singing thing that I could not actually watch because the decoder was demanding data at a rate faster than the hard drive could sustain. I could pause every few seconds to let the disk cache fill up again, but that was hardly a workaround.

By the time 2007 rolled around, and to the chagrin of some who insisted that no one convert their HD subs, there were HD/SD dual-releases mirroring analog/digital television co-broadcasting. Elsewhere, it seemed that every LCD or plasma television came with a ATSC tuner. The local cable co was trying to get us to pony up for a HD set-top box and HD specialty channels. Even my parents were talking about contrast ratios and 1080i. The writing was on the wall.

Team Fortress 2 was coming out. Yes, that's what all the signs were telling me.

At a time when it seemed that HD was finally leaving the station, AMD K8 and Intel Core processors made HD in H.264 possible, with DXVA playing a minor role. DXVA required Windows Vista or better, and if you were unfortunate enough to have Vista (me), you should have had at least a mid-tier Core capable of H.264 720p @ 24fps without any assistance.

Prosumers

For many like myself, the intersection of processing power and fansubs was simply to make fansubs playable in the first place. For fansubbers, more cores or instructions per clock has not fundamentally changed the way they go about their business. Fansubbing continues to be a human-intensive exercise, and turnaround time is more often than not determined by the ability and dedication of the staff.

The most CPU-intensive stage has gotten faster, but there are so many other variables — translator proficiency, difficulty of a show's dialog, number of signs, willingness to cut corners — that encoding may be a bottleneck one episode and not a bottleneck the next. If we were to extend our scope to things like concert video and PV's, video that doesn't need translation, then the chief benefit of a beefier processor is that visual quality improves while encoding time and encoder skill remain constant.

What has really changed is the mass acceptance of soft-subtitles amongst fansub watchers and the proliferation of soft-sub only releases. Screen cappers are dancing in the streets, but this also means that the subtitle script is up for grabs. Re-muxing has become standard practice for "archival" groups that have access to DVD or Blue-Ray raws but either lack a translator, or would prefer to not start from scratch and modify an existing translation instead.

It has also been used in more provocative ways, like re-muxing for currently airing series to make a statement or just to show it can be done with impunity. CrunchyRoll's offerings are fair game, and a frequent target of those who simply dislike CR, dislike the quality of the translations, dislike the quality of the video, or some combination of the three.

One potential application of re-muxing is the development of a subtitle repository in the same vein as OpenSubtitles. However, I wouldn't count on this actually happening, ever, for a whole host of reasons that have nothing to do with technology.

P.S.

If Apsalus had a Core 2 Duo, it might have been completed in time to actually make a difference.

How we measure time and subsequently bundle it up into countable units is pretty arbitrary. Arbitrarily precise, maybe, but it still required a bunch of people to collectively say "make it so."

The same is true on a more macroscopic level; people just decided at some particular point to begin counting the number of times the Earth has orbited the sun, even if they didn't know it then. That combined with the fact that most of us have ten fingers means that we are on the cusp of a year whose decimal representation has some numerically pleasing properties.

Optimistically speaking I have just over seven years of perspective, but honestly it's more like two. So I'm almost certain to put foot in mouth were I to do what the cool kids are doing, and I have yet to acquire a taste in sock, even if it does have a nose of fresh lemon. I hope that I will have more confidence to do a far-seeing, in retrospective, hindsight thing at the end of 2020.

But before this arbitrary year is arbitrarily finished and arbitrarily tossed, I want to raise some awareness about how technology has influenced fansubbing, starting with bandwidth. I don't aspire to any greater goal than awareness: there will be little in the way of dates or citations, filled with handwaving claims the majority of which are meritless.

Then again, all the cool kids are doing it.

Distro: The Real XDCC Blues

I had some brushes with anime in 2000. There was the very first pilot episode of Gundam Wing on YTV that made me an instant TWO-MIX fan (but not a GWing fan) even though I didn't know what they were called or understand what was being said. And there was Fox's "interpretation" of Escaflowne which made me a … non-fan. Meanwhile at school, a couple people were raving about Cowboy Bebop because it had jazz background music.

Hypothetically speaking, if they had bought the OST, those would be the only jazz CD's they would ever own. Likewise, if they had mp3's instead, those would be the only jazz mp3's in their playlist.

Fast forward to some time late 2001 or very early 2002. Cowboy Bebop: Knockin' on Heaven's Door was now fansubbed, and my friend described to me how he had stayed up till 4 in the morning, sitting in an IRC channel with thousands of others, waiting for the bots to be prepped.

I'm not even sure if IRC channels back then could support thousands of users, but we'll run with it.

Remarkably, the whole thing was downloaded in several hours. He had a cable modem, just like everyone else. As far back as 2000, and maybe even 1999, high-speed internet was available in at least the metropolitan areas of the US and Canada. Gamers began to stop complaining about low-ping bastards and started crying about server lag and choke instead. Few batted an eyelash at 300 kbps downlink. The bandwidth was there; people just had to figure out what to do with it.

And what of the file host? Most likely, and by hook or by crook, it was on some university or college optical backbone. Which makes sense, most fansubbers being of that age. But this kind of client-server distribution model could not be sustained by campus students with traffic quotas, or r00ted b0x3n eventually shut down by IT. BitTorrent was the go-to distro solution for precisely those reasons.

Despite the rise of BitTorrent, distro through XDCC was never abandoned and if anything, has experienced a resurgence. Even as media companies began to track the IP addresses of BitTorrent users, bandwidth was getting cheaper, making it affordable to simply install XDCC bots on rented servers connected to 100 Mbit/s pipes. Although the bot's IP address could be discovered, it may be located in an area with lax or non-existent copyright enforcement.

Today, it is possible to get over 1 TB of disk space on a server with a Gigabit Ethernet connection for under $100 / month, with no traffic quota. Depending on the popularity of the group, several bots of this capacity can be funded on donations alone.

But service at this price point has also led others to explore other client-server models: direct download (DDL) sites are a popular alternative to XDCC, while streaming subtitled anime has become viable for small or medium scale entities, on both sides of the legal divide.

Capping RAW

Fansubbers have to get their video from somewhere, and the term "raw" came to denote the video that had yet to be used by the fansubbing group's encoder. Circa 2000, the two primary raw sources were analog television recordings, and DVD's. Analog television was digitized by TV tuner cards, and DVD raws were either imported or downloaded.

Capacity issues at the time made sending minimally processed analog video infeasible, and so the person doing the capturing (the "raw capper") typically performed analog artifact removal, inverse-telecining, and video compression. It was also not unusual to see pre-compressed DVD video as well. In some sense, raws were not really raw, they were marinated.

Most fansubbing groups faced pressure from an enthusiastic and growing viewer base to deliver a subtitled episode in roughly two weeks following the broadcast date, too soon for DVD's to be made available. This placed them at the mercy of raw cappers, and those who were quick and capable in handling analog video gained a following.

Things began to change in 2003 with the introduction of ISDB-T, Japan's terrestrial digital television broadcast standard. With the audio and video "efficiently" packed into an MPEG-2 transport stream, the raw capper was technically obsolete, allowing a fansub encoder a degree of control that was previously only available with raws from DVD ISO or digital satellite TV.

Except for streamed OVA's, short turnaround fansubs of series airing today are typically derived from transport streams. Much of the traditional duties performed by raw cappers have now shifted to the encoder, but the current state of anime production and video processing tools means that transport stream processing tends not to be a labour intensive process. However, pre-compressed broadcast raws continue to be distributed for raw watchers and fansubbing groups that lack the will, means, or personnel to include an encoding step in their workflows.

The explosion in internet capacity has also made the distribution of DVD ISO's widespread, but Blue-Ray ISO's are so uncommon as to be effectively non-existent. With a current 50 GB upper limit, Blue-Ray ISO's will likely be out of reach for all but encoders and the most dedicated fans for some time, even if the industry moves to embrace Blue-Ray releases.

Baked goods

Japan's broadcast television industry, like the US and Canada, maintains a 1080i transmission policy. The result is that anime over digital terrestrial television is interlaced footage with a resolution of either 1920x1080 or 1440x1080. Yet the move to high-definition resolutions for fansubs was initially resisted.

A standard definition fansub, encoded with XviD / mp3 and packaged in an AVI container, was typically targeted for a 175 MB file size. The first de facto HD target was H.264 / mp3 in MKV at 233 MB. There was a small but vocal backlash from those who not only lacked the appropriate processing power and display resolution, but also objected to the 33% increase in file size, a figure that quickly went to 100% with the advent of 350 MB fansubs.

Nevertheless, 1280x720 became entrenched, although for a period of time there were SD versions made available through dual-release or by volunteer/disgruntled re-encoders. Today, SD fansubs are more or less dead, and the community is in the early stages of migrating to 1080p.

With current compression technology, 1080p represents a 60% increase in size above 720p. However, storage, processing power, and bandwidth are all there to support mass adoption of 1080p right now. The primary barrier is impatience: are viewers willing to wait that much longer? I would speculate the answer is no, and throughput will have to catch up first.