Encoding on Linux - a few hints

[lachs0r]

I've been doing this for quite a while now and I thought I should share my knowledge with everyone here.

When trying to encode fansubs on Linux, there are a few issues:

• No (working) native AviSynth
• VirtualDub/DirectShow doesn't work too well with wine
• No native subtitling filter that works like VSFilter (and can emulate it's bugs correctly)

However, there's a tool to solve these problems:
Avidemux (and avsproxy.exe, which comes with Windows builds of Avidemux).

Instructions:

1. Start Avidemux (the native version, of course)
2. Execute wine avsproxy.exe yourscript.avs in a terminal
3. When avsproxy is ready (there should be no errors and a message "Waiting for client to connect... "), in Avidemux, select Avidemux->Connect to avsproxy
4. Now you can encode using your preferred settings

There are some things to consider when writing your scripts:

• DirectShowSource didn't always work the last time I've checked (about 8 months ago), so you should use FFmpegSource2 instead (which seems to do a better job with converting VFR sources to CFR anyway)
• There's no sound output, you have to use a prepared external .wav for Avidemux
• Only YV12 color space is supported for video output (but that shouldn't be a problem, because MPEG4-ASP and MPEG4-AVC use this color space anyway)

The good news is that everything else seems to work fine without modifications, including VSFilter.

I hope that helps a few people here, feel free to start the usual Linux vs. Windows trollfest.

[zhurai-tsuki]

hmm nice, thanks :O

[martino]

OK, I didn't encode for nearly a year now... but I managed to avoid the whole AviDemux stage and just got around with VDub, AviSynth, avs2yuv and BePipe perfectly under wine (DGIndex and YATTA worked too for those that care), the remaining apps were all native (x264, oggenc, ffmpeg, you name it). Also DSS worked just fine, 98% of the time anyway, as did sound output. All that was needed was a native quartz.dll to be registered and for the codecs to be installed, IIRC. Mind you, it's been a long time since I did this the last time...

[zhurai-tsuki]

Originally Posted by martino

OK, I didn't encode for nearly a year now... but I managed to avoid the whole AviDemux stage and just got around with VDub, AviSynth, avs2yuv and BePipe perfectly under wine (DGIndex and YATTA worked too for those that care), the remaining apps were all native (x264, oggenc, ffmpeg, you name it). Also DSS worked just fine, 98% of the time anyway, as did sound output. All that was needed was a native quartz.dll to be registered, IIRC. Mind you, it's been a long time since I did this the last time...

what version of wine was it back then o.o

[martino]

1.1.8/9 or something was the latest I remember using for encoding. Never run AviSynth after that, so no clue how well (or how badly) it works now compared to back then ^^.

[Emess]

Originally Posted by lachs0r

I've been doing this for quite a while now and I thought I should share my knowledge with everyone here.

Most people here have been doing it longer than you.

Originally Posted by lachs0r

No (working) native AviSynth

There is, it's just not public, because nobody wants to share and encourage server-side encoding. Personally I agree with that outlook.

Originally Posted by lachs0r

VirtualDub/DirectShow doesn't work too well with wine

Works fine for me in both linux and freebsd, and I use a distro that's known for it's wine issues.

Originally Posted by lachs0r

No native subtitling filter that works like VSFilter (and can emulate it's bugs correctly)

http://forums.animesuki.com/showthread.php?t=85227 Faster than VSFilter too, a lot of people are asking for a dshow filter for libass it's that good.

Originally Posted by lachs0r

Avidemux (and avsproxy.exe, which comes with Windows builds of Avidemux).

Don't ever use avidemux, it's terrible.

Originally Posted by lachs0r

DirectShowSource didn't always work the last time I've checked (about 8 months ago), so you should use FFmpegSource2 instead (which seems to do a better job with converting VFR sources to CFR anyway)

Not only are you doing it wrong, you don't seem to know anything about DSS. DSS automatically assumes and converts to CFR, it is not capable of VFR. ffms2 on the other hand can do whatever. AVS itself is not VFR aware however so it's not a big deal. Neither source plugin converts to CFR, it assumes CFR and creates VFRaC.

Originally Posted by lachs0r

There's no sound output

Google for avs2wav. You shouldn't be using avs for much audio stuff anyway, trims should all be done with split_aud.pl, which runs natively in *nix anyway.

Originally Posted by lachs0r

Only YV12 color space is supported for video output (but that shouldn't be a problem, because MPEG4-ASP and MPEG4-AVC use this color space anyway)

Not even remotely true. I've output RGB perfectly well, as well as I422. x264 and whatnot all require YV12 input anyway so again, not a problem so much as an ignorant user.

Originally Posted by lachs0r

The good news is that everything else seems to work fine without modifications, including VSFilter.

VSFilter has a bug in the way it handles spacing, you might not have noticed it is all. It's wine's fault with VSFilter and no matter your VSFilter interface the bug is still there.

Originally Posted by lachs0r

I hope that helps a few people here, feel free to start the usual Linux vs. Windows trollfest.

Linux is dumb anyway, go use fbsd now.

E~

[lachs0r]

Of course many people here have been encoding/fansubbing longer than me. That however is no reason for this arrogant attitude towards me.

AssRender didn't work for me with different recent wine releases (crashes as soon as subtitles have to be displayed, didn't have time to debug yet), maybe I've just been doing something wrong. Edit: Fixed it, somehow it only detected fonts in my wine prefix, and even Arial was missing there.
Also, try loading some of the bigger scripts like karaoke effects without splitting them into pieces.

Don't ever use avidemux, it's terrible.

What makes it terrible and why shouldn't I use it? Did I miss something?

Neither source plugin converts to CFR, it assumes CFR and creates VFRaC

Not entirely true. Just assuming VFR as CFR (VFRaC, which is the default for both plugins) would lose audio sync in most cases. Duplicating and removing frames for CFR conversion still has to be handled by the source plugin. Also, why should FFMS2 clearly name it "VFR -> CFR conversion" in it's documentation if it isn't even true?
I think I should explain what I was referring to with FFMS2 "doing a better job" at it.
As an example, take a source where one part is 60 FPS, the rest is 24000/1001 FPS.
Now converting FPS to 24000/1001 with DSS sometimes creates laggy movement (it just assumes 60 FPS as 24000/1001 and skips every ~25 frames), which I didn't experience with FFMS2.
But probably I've just been doing it wrong, as you say.

There's no sound output

Only YV12 color space is supported for video output

These are limitations of avsproxy, nothing more.

I didn't experience the spacing bug. No matter what script I throw at VSFilter in wine, the results are the exact same as on Windows. There's only one exception: It doesn't automatically embolden fonts that don't have a bold face available. Maybe there's an option in fonts.conf but meh, it's a bad practice anyway. AssRender should "fix" this. An alternative would be using FontForge, which should create better looking results, but takes five minutes of your precious time.

Linux is dumb anyway, go use fbsd now.

Guess what's running on my router :P
I don't really like it's licensing though, but that's just me being a GPL/LGPL nazi. Whatever.

Thanks for the reply however.

[Emess]

Originally Posted by lachs0r

What makes it terrible and why shouldn't I use it? Did I miss something?

It just does lots of stupid things and doesn't read a lot of sources properly.

Originally Posted by lachs0r

Not entirely true. Just assuming VFR as CFR (VFRaC, which is the default for both plugins) would lose audio sync in most cases. Duplicating and removing frames for CFR conversion still has to be handled by the source plugin. Also, why should FFMS2 clearly name it "VFR -> CFR conversion" in it's documentation if it isn't even true?
I think I should explain what I was referring to with FFMS2 "doing a better job" at it.
As an example, take a source where one part is 60 FPS, the rest is 24000/1001 FPS.
Now converting FPS to 24000/1001 with DSS sometimes creates laggy movement (it just assumes 60 FPS as 24000/1001 and skips every ~25 frames), which I didn't experience with FFMS2.
But probably I've just been doing it wrong, as you say.

I am too lazy to explain in proper and I worded it poorly previously. DSS assumes a framerate unless you tell it something specific, DSS2 converts to one. They both do it somewhat arbitrarily but are not too bad. Try running them on anything with null frames though, it's a mess. ffms2 is the only source plugin that should be used on pretty much everything besides transport and program streams. It handles VFR correctly.

Originally Posted by lachs0r

These are limitations of avsproxy, nothing more.

Should be using avs2yuv for video and avs2wav for audio. It's the closest you'll get to native avisynth support without Knowing People.

Originally Posted by lachs0r

I didn't experience the spacing bug. No matter what script I throw at VSFilter in wine, the results are the exact same as on Windows. There's only one exception: It doesn't automatically embolden fonts that don't have a bold face available. Maybe there's an option in fonts.conf but meh, it's a bad practice anyway. AssRender should "fix" this. An alternative would be using FontForge, which should create better looking results, but takes five minutes of your precious time.

Odd, you are the first person I don't know to have experienced it. Maybe your spacing is non-0 so it didn't effect you, it's only an issue on 0 spacings. As for the bold thing, that's probably a freetype issue more than anything, get into #aegisub on IRC and ask greg about it.

Originally Posted by lachs0r

Guess what's running on my router :P
I don't really like it's licensing though, but that's just me being a GPL/LGPL nazi. Whatever.

Thanks for the reply however.

People prefer GPL to BSDL, whaaaaaaaaat? D:

You're welcome.

E~

[TheFluff]

being someone who has encoded far longer than emess and is far better than him at being an elitist asshole, I cordially ask him to please shut up

twc is a terrible unfunny troll that should get out, but he's actually right on one point, namely that if you don't care about vfr and have a decent decoder/splitter for the format you want to open (ok well that's a lot of ifs), dss2 is actually far less buggy than ffms2, plus faster (no indexing) and easier to use
chew on that one for a while

my libass avs filter is really more of a proof of concept thing than a real filter. I do not recommend using it for actual releases. it's also linked against a libass that is kinda outdated now.

Originally Posted by Emess

As for the bold thing, that's probably a freetype issue more than anything, get into #aegisub on IRC and ask greg about it.

more like it's a windows-specific (mis?)feature


also: people who use the gpl for non-trolling purposes are usually worse than the riaa/mpaa when it comes to making threats about suing people

[Starks]

emess, are you telling me that someone actually bothered to make avisynth cross-platform? is avs3 making a comeback?

[Emess]

No, it is not cross platform, it is very specifically FreeBSD and Gentoo/Arch x86_64. It might work on other things but I have no idea. The code is split from 2.5.8 rather than a complete rewrite, and only certain filters work on it iirc. AVS3 is dead and gone. I am also aware of a new project that is similar to avs3 in parts but also completely different, that may or may not replace avisynth as a whole.

E~

[Starks]

Why isn't that avisynth branch public (does it even have a name?) and what is the name of that new avisynth-like project?

[fireshark]

Originally Posted by Emess

There is, it's just not public, because nobody wants to share and encourage server-side encoding. Personally I agree with that outlook.

Reading is always a good idea.

[Starks]

Is server-side encoding the next big thing for fansubbing? Aside from off-loading a task that previously required a beefy computer, how does it work?

[lachs0r]

It's fairly easy to use another host as frameserver, wether you modify avsproxy/Avidemux/whatever or use avs2yuv + mkfifo + NFS.

Also, the term server-side encoding doesn't have a real meaning at all. It doesn't even make sense in this context, since AviSynth is a frameserver and not really doing any encoding. The encoding is done by a client, such as VirtualDub or Avidemux.
Hence, load distribution would be a better term.

[TheFluff]

When he says server-side he means encoding on a remote box with a fast connection and better CPU than what you have at home. It's exactly like encoding at home except you do it over RDP or VNC, and it gets downloaded, encoded and uploaded faster. Another advantage is that other people can download and mux the encode even after the encoder has gone to bed.

It pretty much has no downsides other than the fact that doing things over RDP is annoying, so yes, it's a good idea if you can afford a server.

[TrickyDream]

I've been using avs2yuv for a while and it works pretty good =D
avs2yuv to mencoder(ffv1 lossless) or x264.
About thd audio, before I made my own script to split the audio stream I was using foobar to convert the audio to wav and then I used to open virtualdub and load an avs with audiodub + audiodelay + trim and "save to wav", and the native neroAac to convert wav to aac.
I couldn't make avs2wav work. But indeed split_aud.pl is the best choice if you don't have to reencode.

[Emess]

Originally Posted by Starks

Why isn't that avisynth branch public (does it even have a name?) and what is the name of that new avisynth-like project?

Because GPL is stupid and it's a private branch, nobody working on it gives a shit about sharing.The other project is as of yet un-named.

Originally Posted by TheFluff

When he says server-side he means encoding on a remote box with a fast connection and better CPU than what you have at home. It's exactly like encoding at home except you do it over RDP or VNC, and it gets downloaded, encoded and uploaded faster. Another advantage is that other people can download and mux the encode even after the encoder has gone to bed.

It pretty much has no downsides other than the fact that doing things over RDP is annoying, so yes, it's a good idea if you can afford a server.

Except that most people I can think of that do it don't bother downloading it to check any whatnot, so half the time it comes out with a borked IVTC or otherwise looking like shit. Commie's Letter Bee for example, broken IVTC everywhere.

E~

[Dark Shikari]

Originally Posted by Emess

Except that most people I can think of that do it don't bother downloading it to check any whatnot, so half the time it comes out with a borked IVTC or otherwise looking like shit. Commie's Letter Bee for example, broken IVTC everywhere.

I'm pretty sure people do that regardless of where they do the encoding.