Alpha tools for USF subtitles, available for testing ....

[Liisachan]

A new build for you, Liisa: PixiShow_20050724-4 (http://corecodec.org/frs/?group_id=19&release_id=136#r136) :D

changelog:

- fixed incorrect pretty-formatting removal routines
- fixed karaoke on ruby clusters
- fixes to the font selection dialog
- simple zoom in-out in font preview
- new font selector dialog

Thanks for your hard work, it works again now. I'm going to test USF, not only Ruby features but everything including SVG, more deeply sometime in the future (I've been busy recently). Karaoke+Ruby rocks, it's one of those things that are impossible or difficult with SSA/ASS.

In my quick test, it crops a too long ruby text, and that is actually an old problem as you can see here:
http://usf.subforge.net/old.php

IIRC unmei and I once talked about it, and I suggested that the renderer should change the letter-spacing ({\fsp} in SSA lingo) for the ruby base if the ruby text is too long (easier to say)

As another note, more ppl from the CJK area might want to test USF, but the problem is, Corecodec.org is not reachable from some parts of Asia (which is an old, known problem too). I always have to use a proxy to access Corecodec.org.

[BetaBoy]

As another note, more ppl from the CJK area might want to test USF, but the problem is, Corecodec.org is not reachable from some parts of Asia (which is an old, known problem too). I always have to use a proxy to access Corecodec.org.

Liisachan.... sorry for any issue releated to connecting to CoreCodec.org. The problem is related to a range of IP's that continue to DOS CC.org on an off. I will work with both Cyt0plas and or ISP to resolve the issue today.

[unmei]

Liisachan, please forget about USFxSVG for a while :) We are a bit busy fixing and improving what should work first before i am seriously going to spend time on that one.

Next is, i have read-direction implemented for a large part. Maybe i will also be able to fix the ruby cut-off problem on the way, since i need to work on the ruby code for r-d anyway. Here i also have a question for you: How do you think "latin" words in top-down layout should work? While languages that use top-down flow have the characters "upright", it seems some (or all?) rotate western words by 90° clockwise then included in-between (a picture) (http://homepage.hispeed.ch/~mieru/usf/pxf_read-direction_sample3.png) Now, should i rotate western letters or should i leave them upright as the character in the main script are? I implemented that in an halfassed way - actually i rotate characters if they have an ordinal value of less than 255. I know i would need more for proper support of accents and umlauts, but i first have to determine how i can recognize what letters to rotate (maybe you can help me on that as well? :) ).
Also, am i wrong in assuming that ruby comes "towards the begin of the text" compared to the base ? That is, in asian right-to-left line progression, the ruby text comes to the right of the ruby base (and to the left for mongolian R-D) (?)

other smaller things i intend to update/fix in the next time include non-integer values for outline/shadow-level, offset .. (other suggestions welcome)

[Liisachan]

Your picture is correct in usual cases.
As an exception, an abbreviation in Latin alphabet, such as "EU" , doesn't usually "rotate." You can see the samples for the both in:
Glyph orientation within a text run: the 'glyph-orientation-vertical' and 'glyph-orientation-horizontal' properties (http://www.w3.org/TR/2003/WD-css3-text-20030226/#GlyphOrientation) (CSS3 module: text, W3C Working Draft 26 February 2003) ; see also
CSS3 module: Ruby (http://www.w3.org/TR/css3-ruby/)

Also, am i wrong in assuming that ruby comes "towards the begin of the text" compared to the base ? That is, in asian right-to-left line progression, the ruby text comes to the right of the ruby base (and to the left for mongolian R-D) (?)

RtoL is an archaic style to write Japanese, and if you used that the ruby text would be most probably RtoL too. Practically no one will use that writing direction today tho. Top-to-Bottom Right-to-Left (tb-rl) is commonly used today too, and in that case the ruby text comes right side of the ruby base, and the text direction is top-to-bottom.

http://usf.subforge.net/images/tb-rl.png

[unmei]

Well, seems i didn't express it properly, i meant (tb-rl) with "asian right-to-left line progression". Thus was i wanted to know is exactly what you show in the picture. Thanks :)

[ac-chan123]

Unmei asked for latine letter writing in a japanese text in tb-rl wirting. On nearly all Magazines a know they use not rotatet latn letter an arabic numbers(e.g for TV, LD Names, ...). The way you have writen "cat" in the picture(rotated 90°) is a really old way and i haven't seen it in any Magazines since years.
But maybee it's only in Anime Magazines(Newtype, Animage, Anime V, ...) not common. ;-)

[Liisachan]

Yeah, both are possible. The style you are talking about is common in newspapers and magazines, especially for an all-capped word (CAT, not cat) as I mentioned about "EU," and even sometimes for a word in lower case.
I guess newspapers use that style for a technical reason, ie they have to use a fixed set of glyphs.

The traditional styling (rotation) for European words is very common in tb-rl styled books (almost all novels, and some of other books).

I don't know much about prosubs in Japanese, but usually they don't use tb-rl style.
In fansubbing in Japanese language, apparently some subbers use tb-rl. See this thread:
http://forum.doom9.org/showthread.php?s=&threadid=55312

One of such subbers even asked clearly this question (http://mion.wisnet.ne.jp/bbs/1/?nobody=1287) in January, 2004:
"I'm trying to sub vertically using SSA, but I don't know how I can rotate glyphs for English words etc. How can I do that?"
There are a few hacks (http://www.faireal.net/articles/7/23/#d30626), especially so-called "@-marked font hack" or "method 2" (+Sylf's suggestions) seems nice (read the above thread in doom9). If you use an @-marked font, half-width glyphs will be rotated automatially, as you can see in this demo on JTrim, and basically this is what natives would expect.

http://usf.subforge.net/images/at-font.png

For more infomation about Full-width and Half-width, check Unicode Standard Annex #11: East Asian Width (http://www.unicode.org/reports/tr11/)

EDIT: 2 pics added (actual examples from id11 OP, L-E/IDE)

http://ssa.subforge.net/vertical/id11a.jpg
http://ssa.subforge.net/vertical/id11b.jpg

[ac-chan123]

About japanese prosubbing. All japan DVDs with subs has a horizontal left to right writing. Also news(eath quake and such thinks) a shown in this writing in japanese TV. The old vertical right to left writing is mostly for Film Title or for special use(on the side of a shown picture a description or his title).

[unmei]

http://homepage.hispeed.ch/~mieru/usf/pxf_rubyv2_RD_asian_karaoke2.png
(note that this is the test app, not any of the released).

Fixed fontstyle in effects. Implemented asian read-direction ruby (no mongolian yet tho) also with karaoke. Ruby text longer than its base is no longer cut off - but there is no modification of the base, it simply generates "overhang" on both sides.

This is compiled into the picsubu of today and kaiousama is going to release a pixishow with it too (today i think, but it is not in the first pixishow release of today).

[Kaiousama]

PixiShow_20050731-2 (http://corecodec.org/frs/?group_id=19&release_id=139#r139) and picSubU 2005-07-31 (http://corecodec.org/frs/?group_id=19&release_id=138#r138) are out with the new features introduced in unmei's last post.

Also a restyled USF website is avaible at http://usf.corecodec.org

:)

[Kaiousama]

USF website (http://usf.corecodec.org) is now avaible in italian language :D

If someone is interested in translating the website in his/her native language, please contact me.

[Liisachan]

I added real samples from id op.
http://ssa.subforge.net/vertical/id11a.jpg

And a few examples of more complicated situations (just for reference):
http://usf.subforge.net/images/tate-chuu-yoko.png
(gits and tmm)

EDIT:
This page (from Photoshop help::Setting options for Asian type (http://etc.agnesscott.edu/Help/1_12_8_0.html) is informative:
Using tate-chuu-yoko (http://etc.agnesscott.edu/Help/1_12_8_3.html)

Tate-chuu-yoko (also called kumimoji and renmoji) is a block of horizontal type laid out within a vertical type line.

[Kaiousama]

New in PixiShow_20050814 (http://corecodec.org/frs/?group_id=19&release_id=143#r143) :

- Fixed missing refresh on apply changes in property pages
- Better property saving on windows registry

effect and picture related fixes:

-first keyframe > 0%
-misc crash (pointer to style server not assigned to styles resulting from keyframe blending)
-was looking for images in whatever happened to be the current directory. Now the directory of the USF file is used.
-actually would output a bitmap telling "not found XY" if a image is not found

[Liisachan]

Tested long (90-minute) subs in USF, with
MKVToolnix 1.5.0 + Haali 20050815 + VSFilter 2.36 + PixiShow_20050814,
on my Windows 2000 SP4.

Result:
(XviD+MP3+USF).mkv = OK
(XviD+MP3+ASS+USF).mkv = "Mkx" Reconnect error: 80040217

What is this error?

[Haali]

Swapping pixishow and vsfilter on the fly doesnt work, and probably won't work in the future.

[Kaiousama]

PixiShow and VSFilter are not allowed to stay in the same filtergraph.

And since PixiShow does not actually support SSA, that subtitle stream cannot be connected to it.

[Liisachan]

too bad ;_;

*sigh*

[unmei]

no, i someone smiling in happiness over a reason to convert SSA to USF on-the-fly :D

[Liisachan]

no, i someone smiling in happiness over a reason to convert SSA to USF on-the-fly :D
Hopefully....
Technically, SSA2USF should be easy unless there are hackish special effects.
in reality tho, i was trying to convert ASS to USF with Chronosub but it's not that easy

reason 1
Chronosub doesnt understand hex, like in
Style: Default,Arial,20,&H80ffcc99,...

reason 2
Chronosub doesnt escape special characters for Actor
example
Dialogue: 0,...,style,Alice & Bob,0000,0000,0000,,This is bad!

reason 3
the output is one long long long line without linebreaks,
hard to read, hard to fix

reason 4
chronosub doesn't take care of {\i1}{\i0} etc

So...kinda long way to go...

[Liisachan]

http://cia.navi.cx/stats/project/ffdshow/.message/6359074

Commit Message
Author: milan_cutka
Project: ffdshow
Module: ffdshow
support for embedded USF subtitles


http://cvs.sourceforge.net/viewcvs.py/ffdshow/ffdshow/src/subtitles/?sortby=date

kudos to milan_cutka :D

[Kaiousama]

new in PixiShow_20050818 (http://corecodec.org/frs/?group_id=19&release_id=144#r144) :

- Embrional support for SSA/ASS/SRT/SUB in Matroska
- New property page for stream format support selection

notes:

SSA/ASS/SRT/SUB are only supported when encapsulated in a Matroska file.

SSA/ASS tags are not supported at the moment.

Regards.

[Skaarj]

- Embrional support for SSA/ASS/SRT/SUB in Matroska

One excellent filter for any subtitles it is good.
But, works only in RGB32 :(

[Kaiousama]

But, works only in RGB32

With the sample files i have almost every colorspace is supported, what kind of features have your videoclips?

[Skaarj]

Last PixiShow for me works only in FFDshow with YVYU and RGB32 output. I tested VP6, Divx, Nero decoders - only color noise.

with PixiShow_20050702 all ok. It is last version which at me normally displays subtitles with any codec and the decoder.

[unmei]

@Liisa-Chan
The message link is no longer available, so i don't know what there would be written.. I was just stumbling around in the CVS you pointed at, and i was there by accident that it uses an absolute and hardcoded path to load the pixiDDL.dll, therefore i tend to think it is not yet ready for use.

About the comma (other thread (http://forum.doom9.org/showthread.php?t=98359&page=2&pp=20)), yeah i'm probably naive :) Anyway, i probably have to fix that even tho i don't like the idea of starting to threat some characters special. Just not now, next days i have to play with a new toy i got (my 1st laptop ever) =)

[Kaiousama]

New in PixiShow_20050826 (http://corecodec.org/frs/?group_id=19&release_id=147#r147) :

- Added support for the following SSA/ASS tags:
\b\i\u\s\fn\fs\fsc\fsp\1a\2a\3a\4a\1c\2c\3c\4c

:)

[Kaiousama]

New in PixiShow_20050826-2 (http://corecodec.org/frs/?group_id=19&release_id=148#r148):

- Added support for the following SSA/ASS tags:
\frx\fry\frz\a\an\k\
- Added support for ASS layers
- better support for badly nested SSA/ASS tags

[Kaiousama]

new in PixiShow_20050827 (http://corecodec.org/frs/?group_id=19&release_id=149#r149):

- Fixed wrong \1a\2a\3a\4a tag parsing
- Added support for the following SSA/ASS tags: \bord\shad
- Fixed behaviour when Default style is called as "*Default" from the dialogue line
- Fixed wrong tag conversion when \b0\i0\s0\u0 have no \b1\i1\s1\u1 before

[Kaiousama]

Last PixiShow for me works only in FFDshow with YVYU and RGB32 output. I tested VP6, Divx, Nero decoders - only color noise.

Is your videclip height mod 32 ?

[E-Male]

i wonder if a spin-off from the PixiSHow code would be possible:
a SSA/ASS to USF converter
since the parser code is already there (though i don't know if the implementation allows it)

[Kaiousama]

i wonder if a spin-off from the PixiSHow code would be possible:
a SSA/ASS to USF converter
since the parser code is already there (though i don't know if the implementation allows it)

Yes it is.
Parser code is ready as well as style conversion and tag conversion from SSA/ASS to USF.

There are differences in dialogue parsing since when encapsulated in matroska container, an SSA/ASS dialogue line is rearranged in order to minimize information redundancy, here (http://www.matroska.org/technical/specs/subtitles/ssa.html) you can find detailed information about this rearrangement.

If anyone is interested in creating such a tool, he/her can contact me for any useful information.

[jjseth]

Hi!
I always play video+embeded subtitles with VSFilter but I was testing the last PixiShow-VSFilter 2.36-HaaliSplitter-MediaPlayerClassic and embeded .srt subtitles wasn't showed and full screen video killed my little 256Mb RAM....
I uninstalled PixiShow but VSFilter don't show any embeded subtitle anymore.
I uninstalled/reinstalled VSFilter, etc but....
VSFilter show external subtitles files OK, the problem is with the embeded subtitles.
MediaPlayerClassic, VSFilter and Haali show the subtitles streams fine but in the video this don't appear.
Any idea?

I don't post this dude in the Subtitles sub-forum because PixiShow is only mentioned here but if any moderator move the post OK! ;)

Edit.
The problem was the Matroska "SimpleBlock" mode. :o

[Liisachan]

- Haali's splitter once had a bad uninstaller, and it didn't unregister splitter's info properly. This could mess up PixiShow-related DS graphs. Manually cleaning the undeleted info about Haali's may fix the problem.

- Apparently PixiShow doesn't like YV color spaces. RGB works fine.

- PixiShow usually causes crashes if "Connect to compatible filters only" is checked in ffdshow.

- PixiShow's merit is higher by 0x1 than VSFilter's. This is needed so that it will be loaded for (video+usf).mkv. But PixiShow wants to handle SSA/ASS too by default. So it is possible that VSFilter doesn't load when desired for the file such as (video+ass).mkv, and in that case the result may be unstable (the player may crash) since PixiShow is alpha after all. Disabling SSA/USF supports in the PixiShow Properties box should fix the problem.

- Apparently PixiShow works fine not only with Haali's splitter but also Gabest's splitter and MPC.

[Rasqual]

Hello, I have trouble making the ruby styles work in USF. Please tell me where I am wrong.

I am using PixyShow 20050827 and MPC 6.4.8.6 (either built-in or Haali's) and read from a Matroska container.

I would like to use these definitions:

<style name="rubyStyle">
<fontstyle
face="MS Gothic"
size="18"
color="#9999ff"
outline-color="#f1f3f8"
outline-level="1"
shadow-level="0"
/>
</style>

<style name="rtStyle">
<fontstyle
size="10"
outline-level="0"
/>
</style>

that is to say apply a style to the ruby text element.

I tried to have it applied to the following subtitle piece:

<subtitle start="00:00:21.900" stop="00:00:28.31">
<text>
<ruby>言葉<rt>ことば</rt></ruby>に　<ruby>出来<rt>でき</rt></ruby>ない<ruby>気持<rt>きも</rt></ruby>ち<ruby>多<rt>おお</rt></ruby>すぎて
</text>
</subtitle>

However, the styles do not cascade nicely as any web designer would expect. I don't think the rtStyle (the name came to me) is read at all. I tried looking at the pixishow/pixifier code (http://cvs.corecodec.org/index.cgi/*checkout*/usf/utils/pixifier/pfRasteriserU.pas?rev=HEAD&content-type=text/plain) (and here (http://cvs.corecodec.org/index.cgi/*checkout*/usf/utils/pixifier/pfImgDescU.pas?rev=HEAD&content-type=text/plain) and here (http://cvs.corecodec.org/index.cgi/*checkout*/usf/utils/pixifier/pfImgDescBuilderU.pas?rev=HEAD&content-type=text/plain)), but I thought you would surely know better where to look at... In fact, I was a bit lost in there.
I think the relevant part is //TO-DO !!
fBaseStyle := builder.makeBaseStyle;
and there is no mention of the ruby text being styled at all.

Not-really-related Note:
Ideally, I am tempted to use the following style, but I don't know whether it is supported and if the scale-* attributes were designed for this purpose. (i.e. scaling implies bilinear/bicubic resize, while specifying a point size uses font kerning, metrics and whatnots). Your opinion?

<style name="rtStyle">
<fontstyle
scale-x="0.55"
scale-y="0.55"
outline-level="0"
/>
</style>


Fianally, regarding the merit of pixyshow, I seem to have trouble building the graph in zoom player (4.51) because VSFilter is chosen instead even though the merit of PS is higher as Liisa-chan pointed out. Meanwhile, graphedit loads the PS filter correctly.
Is there any specific action to configure? PS is not blacklisted and 'Manual Filters' is disabled. Naturally, if I disable VSFilter, PS is correctly included in the graph.

[Liisachan]

Last time I tested it (http://usf.subforge.net/20050625.php), <style name="rubyStyle"> worked, like that style name was hard-coded. You are right, I remember I felt the similar thing too.

[unmei]

in my test app it seems to work, but as i havent worked on it for some time and kaiou also hasnt made PS updates for some time, i'm not exactly sure the test app is in sync with his last released PS.

Anyway, just to make sure, if your ruby style is not called "rubyStyle" you have to mention it in each <rt> element you want it applied like this:

<subtitle start="00:00:21.900" stop="00:00:28.31">
<text>
<ruby><rb>言葉</rb><rt style="rtStyle">ことば</rt></ruby>に　
<ruby><rb>出来</rb><rt style="rtStyle">でき</rt></ruby>ない
<ruby><rb>気持</rb><rt style="rtStyle">きも</rt> </ruby>ち
<ruby><rb>多</rb><rt style="rtStyle">おお</rt></ruby>すぎて
</text>
</subtitle>


Of course this also means you can have different styles for different ruby texts ;)

[edit]
I just realised you had the <rb> elements missing in your snippet - that could well have been the cause for it not working as intended..

[Rasqual]

Yes indeed, the <rb> elements were missing.
Is there a way to avoid repeating style="rtStyle" everytime?
Otherwise, it seems to work fine, although there is still room for improvement regarding ruby overhangs.