[Koe no Katachi](https://myanimelist.net/anime/28851/Koe_no_Katachi)
BlurayDesuYo typesetting (modified), 35mm script (edited to match BDY's typesetting).
The video and audio of this movie are more interesting than usual. Start with the video: this movie has serious banding problems that are exacerbated by how noisy it is. In order to deal with banding artifacts, we encoders generally need to denoise the video first, so if the movie is hard to denoise, it's also hard to deband (without causing all kinds of collateral damage, anyway). For this reason, I was more aggressive than usual with my denoising, substituting a spatial-only denoiser (or rather deblocker) for my normal temporal one in parts of the frame where the temporal denoiser could not find good motion vectors. I also scenefiltered debanding (in other words went scene by scene to see where the banding was especially annoying and adjusted the filtering accordingly), which is the second time I have done so for a project. The previous time I did it was also a KyoAni movie. I don't think this is a coincidence.
To talk about the other scenefiltering I did, I first need to talk about the various sources for this movie. The conventional wisdom appears to be that the Italian Blu-Ray is the best source. It's understandable why—if you look at [comparisons](https://slow.pics/c/YQtSFchu), the Japanese Blu-Ray (JPBD) appears to have a lot of detail wiped compared to the ITABD. But that "detail" we see is *purely* a layer of coarse dynamic noise added by Dynit. Basically, Dynit applied some debanding (perhaps with some denoising beforehand) and then added heavy noise in order to cover up any remaining imperfections. But that's not the only thing the noise covered up—between the JPBD and the ITABD, the former makes it easier to see subtle elements of the frame that are unambiguously detail, like faint colored lineart.
If all Dynit did was deband and add crazy grain, then hell, I can take the JPBD and do that myself. Except I won't—not the grain part, anyway. That's because there are many scenes in this movie where the grain level gets kicked up a notch or two in the JPBD—mostly flashback scenes, but also some others, like at the very beginning of the movie and at 2:04:17. Figuring out the significance of this distinction is an exercise left to the viewer—and the viewer can actually participate in that exercise *because I used the JPBD*. Using the ITABD, with its heavy grain everywhere, makes it somewhat harder to recognize the "grain scenes" and any significance they might have. Anyway, I turned off the denoiser for those scenes.
As a whole, this encode fixes some important problems but is not perfect (to say the least), and when someone makes a comparison to the ITABD and Beatrice-Raws (which in my view are the other two possible candidates for "best encode"), I will link it here. EDIT: https://slow.pics/c/0bVG2tnn
On to the audio. [Hydes's torrent description](https://nyaa.si/view/1465150) was instructive. To summarize, the 5.1 tracks included in the Italian and US Blu-Rays have significantly higher dynamic range than the one included with the JPBD and are likely the way the movie is "meant to be heard." Looking at the center channels for the JPBD and USBD 5.1 tracks, the latter is [pretty clearly the original audio](https://i.imgur.com/op194wQ.png), with the JPBD audio having been run through a DRC filter. Hence, the audio included here is from the USBD. I have also included two stereo tracks: the official track and a custom 2.0 downmix of the USBD 5.1 that has higher dynamic range than the official track. The custom downmix is "quieter," but the point is that you should turn your system volume up to compensate so that when loud scenes come around, they become more impactful. The climactic scenes in this movie involve fireworks, after all—and those fireworks were heavily nerfed in the official track. The loud scenes in the custom downmix are jarring, but if Hydes's torrent description is accurate, it was KyoAni's intention to have large volume differences and thus their intention to jar the viewer. Nevertheless, I've set the official track as default since it's a more "normal" listening experience.
There is an alternate honorifics track in this release. Set your media player to play "enm" language tracks by default to automatically play honorifics tracks.
Please leave feedback in the comments, good or bad.
Please read this short [playback guide](https://gist.github.com/motbob/754c24d5cd381334bb64b93581781a81) if you want to know how to make the video and subtitles of this release look better.
**Anyone wanting to do their own release is free to use any part of this torrent without permission or credit.**