ffdshow tryouts project: Discussion & Development

[martino]

This is perhaps a bit slightly offtopic, however I'd like to know whether anyone has tried to run ffdshow in wine and managed to get huffyuv (yv12) vfw decoding working. It installed fine for me, but I can't change any of the dropdown boxes (yet the registry entries seem to be set on 1) which results in a "couldn't locate a decompressor for fourcc ffds in VDub. Any help would be greatly appreciated. Thanks.

[yesgrey]

I got a question about resampling in ffdshow audio.

I find libavcodec to sound better than libsamplerate.

but was ffdshow updated with the recent updates from March and July ?
http://www.mega-nerd.com/SRC/history.html

would that be doable ?

thanks,

No it wasn't. I have looked now in svn and it's the old code.
I will take a look to see if I could update the code. If I could, I will create the new files and then I will request here some help on how to do it in the svn...

[nautilus7]

clsid was against this update: http://forum.doom9.org/showthread.php?t=138534

[pitch.fr]

home about some crazy packing ? like .7z ?
or some optional component ?

I can pack the Resampler.dll from Reclock to 480k with Winrar, so the diff with the older version wouldn't be 800k ?!

[rickardk]

No it wasn't. I have looked now in svn and it's the old code.
I will take a look to see if I could update the code. If I could, I will create the new files and then I will request here some help on how to do it in the svn...

I hope he reconsider as their is a big diffrence in sq.

[yesgrey]

I have finished the look at the code and it seems to be a very easy thing to do. The difference in sound quality from the old to the new is very big, especially using the best mode. I hope clsid reconsiders the change. I understand his point of view, but only if the quality diference was small.
He could always test reclock with the new resampler and test for himself if the quality improvement justifies the change...

[Liisachan]

@pitch.fr
btw did you read this recent post of Inventive Software?
BTW, I think libmad MP3 decoding is partially broken in ffdshow. I'm on 2073 IIRC, and I have a 44100Hz 16-bit file in an AVI that was resampled from 48000Hz 16-bit in VirtualDub. It exhibits clicks and pops alongside the normal audio, but the built-in Windows DShow Decoder (Fraunhofer IIRC) plays it back without problems.

see? libmad is not really reliable, it's as broken as hell in mplayer too, and the worst part of this is there's practically no one, especially no original libmad developer, to fix any bug...

And this is ffdshow's default mp3 decoder...:confused:

[tal.aloni]

Hello Albain,
first, thanks for all of your your efforts.
I waited a while for a free E-AC3 Codec, and I'm glad it's finally here. (ffdshow is great in general)

Just making sure you know, I'm currently using rev 2124 beta3, and the channel mapping problem is still present in 5.1 sources. (libavcodec)

unfortunately, swapping channels isn't a real workaround.
(since other sources doesn't have the wrong input mapping in the first place)

Note:
1. the problem is gone when liba52 is doing the decoding.
2. the problem does not exit when utilizing libavcodec using eac3to.

I can provide some AC3 / E-AC3 samples if needed (dolby setup tests)

Thanks,
Tal Aloni

[nautilus7]

2. the problem does not exit when utilizing libavcodec using eac3to.

Wrong. Libav does outputs channels with incorrect order and eac3to does it's own remapping after the decoding.

It's a general problem of the libav decoder.

[tal.aloni]

Wrong. Libav does outputs channels with incorrect order and eac3to does it's own remapping after the decoding.

It's a general problem of the libav decoder.

Thanks for the clarification.

Tal Aloni

[CiNcH]

Hey guys, is someone into MP3 parsing/decoding? We have implemented MP3 support into our splitter. What we do is splitting the MP3 Elementary Stream into 2K IMediaSamples and send them downstream. Now ffdshow fails to reconstruct the MP3 frames on some VBR files and introduces hiccups. It works properly with the Fraunhofer MP3 decoder coming with MS Windows.
MPA Audio Decoder shows the same effect as certain things within the OpenSource community seem to have spread ;) .

We then implemented a frame parser into the splitter and send them downstream which works around the problem. But this introduces a major flaw into our design.

Conclusion is that ffdshow tries to rebuild frames but fails somewhere, the decoder itself seems to work properly. Is it possible that the padding field within the MP3 header is not evaluated? Padding bytes are especially common with 44.1 kHz files where the encoder sometimes "fails" to meet frame length.

If you like I could provide player + splitter + sample. Just tell me who is in charge of that and is willing to solve the issue. I would be more than happy to contribute.


BTW, I am currently working with Haruhiko Yamagata to solve the NALU/frame thing for H.264/VC-1. A parser is indeed missing in this case only supporting Haali and MPC splitters..

[albain]

Still no sound with: cine.premier.hd.astra.19e.05.feb.2008.ts.
Dca is crashing with: x264.dts.sample.mkv
http://rapidshare.com/files/144998756/samples.7z.html

Crashing problem was due to a bad linking. It is working now.

Concerning your no sound issue, this is quite a problem because on this AC3 2 channels stream the parser seems to clean all the stream before feeding the decoder...

[albain]

If I disable the parser for 2 channels stream, it is working. :confused: weird

The parser can parse only multichannel streams ? (at least for AC3)

EDIT : actually, the parser works great for some 2 channels AC3 streams (Test_LibavAC3.avi), but not for some others (cine.premier.hd.astra.19e.05.feb.2008.ts from xxl)

By the way it would be great to have the channel remapping code if available (I don't have any HT receiver to make this fix)

[madshi]

By the way it would be great to have the channel remapping code if available (I don't have any HT receiver to make this fix)
Here's what eac3to is doing:

procedure RemapAc3ChannelsLibav;
begin
case audioDetails.Channels of
3 : if audioDetails.ChannelStr = '3/0' then
source.AddChildUpStream(ChannelMapper('0,2,1,3,4,5'));
4 : if audioDetails.ChannelStr = '2/1.1' then
source.AddChildUpStream(ChannelMapper('0,1,3,2,4,5'))
else
if audioDetails.ChannelStr = '3/0.1' then
source.AddChildUpStream(ChannelMapper('0,2,1,3,4,5'))
else
if audioDetails.ChannelStr = '3/1' then
source.AddChildUpStream(ChannelMapper('0,2,1,3,4,5'));
5 : if audioDetails.ChannelStr = '2/2.1' then
source.AddChildUpStream(ChannelMapper('0,1,4,2,3,5'))
else
if audioDetails.ChannelStr = '3/1.1' then
source.AddChildUpStream(ChannelMapper('0,2,1,4,3,5'))
else
if audioDetails.ChannelStr = '5.0' then
source.AddChildUpStream(ChannelMapper('0,2,1,3,4,5'));
6 : source.AddChildUpStream(ChannelMapper('0,2,1,5,3,4'));
end;
end;

procedure RemapDtsChannelsLibav;
begin
case audioDetails.Channels of
3 : if audioDetails.ChannelStr = '3.0' then
source.AddChildUpStream(ChannelMapper('1,2,0,3,4,5'));
4 : if audioDetails.ChannelStr = '2/1.1' then
source.AddChildUpStream(ChannelMapper('0,1,3,2,4,5'))
else
if audioDetails.ChannelStr = '3.1' then
source.AddChildUpStream(ChannelMapper('1,2,0,3,4,5'))
else
if audioDetails.ChannelStr = '3/1' then
source.AddChildUpStream(ChannelMapper('1,2,0,3,4,5'));
5 : if audioDetails.ChannelStr = '2/2.1' then
source.AddChildUpStream(ChannelMapper('0,1,4,2,3,5'))
else
if audioDetails.ChannelStr = '3/1.1' then
source.AddChildUpStream(ChannelMapper('1,2,0,4,3,5'))
else
if audioDetails.ChannelStr = '5.0' then
source.AddChildUpStream(ChannelMapper('1,2,0,3,4,5'));
6 : if audioDetails.ChannelStr = '5.1' then
source.AddChildUpStream(ChannelMapper('1,2,0,5,3,4'))
else
source.AddChildUpStream(ChannelMapper('1,2,0,3,4,5'));
7 : source.AddChildUpStream(ChannelMapper('1,2,0,5,3,4'));
8 : source.AddChildUpStream(ChannelMapper('1,2,0,5,3,4'));
end;
end;
This is for a quite old libav build, but I guess it should still be correct for the latest build, too. Channel remapping for AC3 and E-AC3 should be identical.

[nautilus7]

Concerning your no sound issue, this is quite a problem because on this AC3 2 channels stream the parser seems to clean all the stream before feeding the decoder...
I demuxed the ac3 track from the "cine.premier.hd.astra.19e.05.feb.2008.ts" file with eac3to. The ac3 plays just fine.

I have to say that some of these samples that produce problems are "weird". I can't tell if this could be the problem here, but eac3to reports a delay of 7secs for this .ts file which sounds uncommon for a cut sample. :confused:

[albain]

Thank you madshi, I will try to implement your code

Concerning this sample, I agree that the sample may be weird, but the libavcodec parser seems be too buggy too : it wipes out too many bytes, whereas this file doesn't need to be parsed at all (works great if directly passed to the decoder)

[albain]

Crashing on the DTS file occurs in ff_imdct_half_sse function.

I had the same error a while ago and it was fixed by installing a previous version of binutils, but here it goes again

[_xxl]

Crashing on the DTS file occurs in ff_imdct_half_sse function.

I had the same error a while ago and it was fixed by installing a previous version of binutils, but here it goes again

I'm using previous version of binutils. Can you share your mingw environment?

[albain]

I'm using previous version of binutils. Can you share your mingw environment?

MSYS 1.0.11(0.46/3/2) 2007-01-12 12:05 i686 unknown; targ=MINGW32
Target: mingw32
Configured with: ../gcc-4.2.4/configure --prefix=/mingw --host=mingw32 --target=mingw32 --build=mingw32 --enable-languages=c,ada,c++,fortran,objc,obj-c++ --disable-nls --disable-win32-registry --enable-libgomp --disable-werror --enable-threads --disable-symvers --enable-cxx-flags=-fno-function-sections -fno-data-sections --enable-fully-dynamic-string --enable-sjlj-exceptions
Thread model: win32
gcc version 4.2.4 (TDM-1 for MinGW)
binutils-2.17.50-20060824-1

[albain]

The file also crash with mplayer (and also plays fine with no optimization) :
Opening audio decoder: [ffmpeg] FFmpeg/libavcodec audio decoders


MPlayer interrupted by signal 11 in module: init_audio_codec
- MPlayer crashed by bad usage of CPU/FPU/RAM.
Recompile MPlayer with --enable-debug and make a 'gdb' backtrace and
disassembly. Details in DOCS/HTML/en/bugreports_what.html#bugreports_crash.
- MPlayer crashed. This shouldn't happen.
It can be a bug in the MPlayer code _or_ in your drivers _or_ in your
gcc version. If you think it's MPlayer's fault, please read
DOCS/HTML/en/bugreports.html and follow the instructions there. We can't and
won't help unless you provide this information when reporting a possible bug

[_xxl]

binutils-2.17.50-20060824-1
I was using binutils-2.17.50-20070129-1.
Thanks!

[albain]

Here is a new build with I hope the wrong channels mapping fixed.

If this is not the case, please tell me the expected mapping, and the actual one.

http://rapidshare.com/files/145520560/ffdshow_rev2131_20080915_dbt_beta4.exe.html

Concerning the crashing on DTS, this is definetely a compilation problem (binutils/gcc version) or a ffmpeg bug.
I tried binutils 2.17 & 2.18, same problem.
Disabling mmx optimization solves the problem.

[tal.aloni]

Here is a new build with I hope the wrong channels mapping fixed.


Hey Albain,
I'm glad you're working on the problem,
I've tested you're latest build, but I still have problems with mappings:

5.1 channels:
Front Left Channel (1) > Front Left Speaker (1)
Front Right Channel (2) > Front Right Speaker (2)
Center Channel (3) > Surround Left (5)
LFE Channel (4) > Surround Right Speaker (6)
Surround Left Channel (5) > Center Speaker (3)
Surround Right Channel (6) > LFE Speaker (4)

Thanks a lot,
Tal Aloni

[nautilus7]

Yes, I confirm the above wrong mapping for both AC3 and E-AC3.

It was L,RS,C,R,LS,LFE
It is now L,R,LS,RS,C,LFE
It should be L,R,C,LFE,LS,RS

[Inventive Software]

Hey Albain,
I'm glad you're working on the problem,
I've tested you're latest build, but I still have problems with mappings:

5.1 channels:
Front Left Channel (1) > Front Left Speaker (1)
Front Right Channel (2) > Front Right Speaker (2)
Center Channel (3) > Surround Left (5)
LFE Channel (4) > Surround Right Speaker (6)
Surround Left Channel (5) > Center Speaker (3)
Surround Right Channel (6) > LFE Speaker (4)

Thanks a lot,
Tal Aloni

The mapping on the left is for WAV only, the one on the right is for AC-3.

[nautilus7]

What do you mean by this?

Every decoder should output the same channel order (this the "wav" channel order), or else we 'll have to change the speaker's configuration each time we listen to a different audio format. We don't care what is the order of the channels inside the tracks. We just need them to be decoded in the same order.

[Zep]

Hi I just installed ffdshow_rev2110_20080908_clsid.exe and I have a few questions about subtitles.


for ffdshow subs to work I have to

Subtitle-Vobsob-enabled correct? I ask because I can't get subs to work other wise and it seem redundant to have.
Subtitle-enabled then have to have also have Subtitle-Vobsob-enabled unless you can turn off vobsub and still get subs working. I need both enabled?


With both enabled because that is only way I can see subs I am unable to change the rendering in 2 areas no matter what I try. the areas being

Subtitle-text
Subtitle-font

nothing changes. I even apply then restart MPC after each change to make sure but still nothing changes. I tried other players and same thing happens. So I do no think it is a MPC thing but I'm not 100% sure.

I do not have the external vobsub older stand alone or the VSFilter.dll as I only want this handled by ffdshow.

Subtitles-Vobsub area everything does work there but I want to change font and text formats as well. Right now I am stuck with white only subs in Arial. Am I doing something wrong or missing a step?



Thanks

[Anima123]

When enable dolby decoder, it still affect with plain stereo audio, which could get unwanted sound if the stereo audio track already does rear channels delay when convert from original 5.1 track. I wonder if we can add an option to make it not affect when rear channels are not fed.

Note: I've got only stereo speaker setup, thus no need to change stereo track up to 5.1 channels.

[Snowknight26]

I only just recently noticed that ffdshow doesn't include any decoders for WMA lossless. Are there any decoders that can do it (I can only dream of libavcodec) that could perhaps be included in ffdshow?

[ACrowley]

I only just recently noticed that ffdshow doesn't include any decoders for WMA lossless. Are there any decoders that can do it (I can only dream of libavcodec) that could perhaps be included in ffdshow?

WMAudiodecoder DMO can handle WMA 9 lossless ,but not ffdshow/libavcodec
But you can process any Audio with ffdshowaudio processor "behind" WMAudioDecoder

[Inventive Software]

I only just recently noticed that ffdshow doesn't include any decoders for WMA lossless. Are there any decoders that can do it (I can only dream of libavcodec) that could perhaps be included in ffdshow?

Funny you should mention that, I posted something on the ffmpeg-devel mailing list asking much the same thing, and the response I got was "if you know or can get the bitstream specifications, we can implement it". MS has the source code licensing for $5000. :)

What do you mean by this?

Every decoder should output the same channel order (this the "wav" channel order), or else we 'll have to change the speaker's configuration each time we listen to a different audio format. We don't care what is the order of the channels inside the tracks. We just need them to be decoded in the same order.

I wasn't aware it worked like that. All I know is that LFE 4th as channel is WAV, 6th properly is Dolby Digital.

[clsid]

for ffdshow subs to work I have to

Subtitle-Vobsob-enabled correct? I ask because I can't get subs to work other wise and it seem redundant to have.
Subtitle-enabled then have to have also have Subtitle-Vobsob-enabled unless you can turn off vobsub and still get subs working. I need both enabled?


With both enabled because that is only way I can see subs I am unable to change the rendering in 2 areas no matter what I try. the areas being

Subtitle-text
Subtitle-font
There are different kinds of subtitles. Text based and image based (VobSubs). It is a separate option because vobsubs do not always work properly with ffdshow.

The text/font options only apply to text based subtitle formats (like SRT) and not to vobsubs.

[pitch.fr]

I hope he reconsider as their is a big diffrence in sq.
so what do you say CLSID ?
you can compare Reclock 1.7b4 and 1.809 to witness the huge quality improvement.

maybe using some 7z/RAR packing could fix the size issue ?

I can compress the new Reclock DLL to 400k with winRAR.

Thanks,

[Snowknight26]

Funny you should mention that, I posted something on the ffmpeg-devel mailing list asking much the same thing, and the response I got was "if you know or can get the bitstream specifications, we can implement it". MS has the source code licensing for $5000. :)

So I guess the foobar2000 devs must have coughed up that much sometime or another. :p

[Zep]

There are different kinds of subtitles. Text based and image based (VobSubs). It is a separate option because vobsubs do not always work properly with ffdshow.

The text/font options only apply to text based subtitle formats (like SRT) and not to vobsubs.

ok I am beginning to understand lol

Anyway, I see the 2 formats in the avi dir are

.idx
.sub

how do I go about changing font and font collor and font size then? I see in the .idx many things that look tweakable haha but surely there is a better way than to change this stuff?



thanks

[clsid]

.idx/.sub are VobSubs. You can't change the font.

[pitch.fr]

@clsid : should I take your silence as a "no" ?

[albain]

Finally, the channel order seems to be correct for AC3/EAC3/Vorbis/AAC multichannel streams using libavcodec

Here is a test build :
http://rapidshare.com/files/146146686/ffdshow_rev2131_20080915_dbt_beta5.exe.html

[Anima123]

AAC decoder still won't work with albain's beta5.

[Zep]

.idx/.sub are VobSubs. You can't change the font.

wow that sucks. I guess I am stuck installing the old external vobsub since at least I can change the font color and stretch/squish the font with that version.


thanks

[rack04]

AAC decoder still won't work with albain's beta5.

Works here.

[Kurtnoise]

.idx/.sub are VobSubs. You can't change the font.
you are wrong here...you can change the font and/or colors from vobsub files as well with SubtitleCreator.

AAC decoder still won't work with albain's beta5.
As I said earlier, the AAC decoder from lavc (libavcodec) supports only LC streams, not HE or PS for the moment.

[Anima123]

I've tried lc-aac audio wrapped in mp4 container but failed to play. Definitely not working for me. :(

[nautilus7]

Finally, the channel order seems to be correct for AC3/EAC3/Vorbis/AAC multichannel streams using libavcodec

Here is a test build :
http://rapidshare.com/files/146146686/ffdshow_rev2131_20080915_dbt_beta5.exe.htmlThanks it works with AC3 and E-AC3 which i tested.

The libav decoder still crashes with every DTS track i try.

[clsid]

you are wrong here...you can change the font and/or colors from vobsub files as well with SubtitleCreator.
Sure, everything is possible with enough effort. But it would require a conversion step to change the font.

[_xxl]

The libav decoder still crashes with every DTS track i try.
Has this been reported to ffmpeg dev team?

[Kurtnoise]

Sure, everything is possible with enough effort. But it would require a conversion step to change the font.
what do you mean by a conversion step ? Just load your file, choose a font/color and save. That's all...:)

[clsid]

I doubt that it is a bug in FFmpeg. Wasn't it a binutils bug? Just disable whatever function is crashing.

[Kurtnoise]

I've tried lc-aac audio wrapped in mp4 container but failed to play. Definitely not working for me. :(
I'll try to do some tests on my side...

[Leak]

what do you mean by a conversion step?
DVD subs (aka .idx/.sub) are bitmaps. If you want to change the font, you'll have to OCR all subtitle images, possibly re-position them and save them in some other format or create new subtitle images.

I think you'll agree that this is a wee bit more complicated than "just changing a setting"...

np: Soul 223 - Q (Soul Jazz Records Singles 2006-2007 (Disc 1))