ffdshow tryouts project: Discussion & Development

[Leak]

I see a strange effect, when I switch to german language in ffdshow. After restarting the player and displaying ffdshow's filter properties, the property window has shrunk on the right side and many items are no longer acessible (window is not resizable). After switching back to english language and restarting the player again, everything is allright.
BTW: Many thanks for Your marvellous work !
What OS are you using? And could you perhaps post a screenshot of the distorted window?

[Leak]

A release is needed that is compatible with all the receivers, it doesn't have sense install a release for every model of receiver.
So how is ffdshow supposed to automagically know that you have a receiver that can't handle high bitrate DTS streams?

[cc979]

@Yong: osd seems broke on cpu load shows svn 1455

weird thing is no text appears when opening filter - the tabs are blank but the details in tab are ok

[haruhiko_yamagata]

I have fixed the fast rendering problem at rev 1457.
At rev 1456, OSD CPU usage is significantly lowed compared to rev 1455, but still a bit slower than before.

@cc979: weird thing is no text appears when opening filter - the tabs are blank but the details in tab are okI don't understand. What do you mean by "filter"?

[DMD]

So how is ffdshow supposed to automagically know that you have a receiver that can't handle high bitrate DTS streams?
I use a sintoamply Yamaha RX795a

[Leak]

I use a sintoamply Yamaha RX795a
That didn't answer my question at all.

How is ffdshow supposed to know what receiver you have, and what it's capabilities are? SPDIF out is a one-way connection, there's nothing coming back from the receiver that identifies it.

Also, you don't have to install a different ffdshow release depending on your receiver, but you just might have to check the "Max. SPDIF compatibility" checkbox - is that too much to ask?

[clsid]

Maybe the option should be changed. Ensure maximum compatibility by default (old behaviour) and require to check a box to enable support for high bitrate DTS. The checkbox could be called something like "high bitrate DTS over SPDIF workaround", with a tooltip like "may not work with some receivers".

[santiago_kq]

Is there any way I can use the *same* FIR filter that ffdshow uses, but with Mplayer / Mencoder on Linux?

I need to batch convert some videos, but when I bump up the volume (needed for my conversion), I need to compensate for the added noise, using a filter... and FIR was the only one that worked great for me, automatically.

Thanks in advance,

[cc979]

@haruhiko_yamagata here is a picture of the bug

http://img521.imageshack.us/img521/8317/clipboard1bu5.png (http://imageshack.us)

[Kado]

@cc979

Isn't it Windowblinds that just messed around with the text (I suppose you're using it)? I'm using the latest build (1463) and everything is ok.

[cc979]

just plain windows, no visual extras

[ShadowVlican]

Maybe the option should be changed. Ensure maximum compatibility by default (old behaviour) and require to check a box to enable support for high bitrate DTS. The checkbox could be called something like "high bitrate DTS over SPDIF workaround", with a tooltip like "may not work with some receivers".
i second this request

that and make the checkbox independent from the AC3 output (people that want AC3/DTS passthrough do not check this box anyways)

[Ginsonic]

What OS are you using? And could you perhaps post a screenshot of the distorted window?

Yes of course ! I am using WindowsXP german and here are the two screenshots (english and german language setting).

[albain]

Exactly.

Rev 1381 works correctly, but doesn't support high-bitrate DTS streams.

Rev 1437 supports high-bitrate DTS streams, but this makes changes to the SPDIF signal which some receivers can't handle.

Therefore, rev 1443 contains a checkbox (maximise SPDIF compatibility) which allows you to toggle between the original behaviour (which doesn't support high-bitrate DTS streams) and the updated behaviour (which doesn't work with all receivers).

Does that make sense?

Ok I am the one who implemented the new SPDIF format, and I didn't figure out why it doesn't work on some receivers. On my receiver every signals are working (high bitrates DTS and any DTS/AC3 stream).
I think that the problem is related to the data buffer which is not filled correctly.
As long as the solution is not found, it is necessary to keep this checkbox in order to switch between the old and the new signal.

Anyway, as proposed clsid, it is a good idea to contact ac3filter developer in order to know how he made a 100% successful SPDIF signal. I checked his code but it's not the same method

If someone wants to dig around :
Buffer size :
old behaviour :
pProperties->cBuffers=4;
new behaviour :
pProperties->cBuffers=10;


Buffer filling :
old behaviour :
pDataOutW[0]=0xf872;
pDataOutW[1]=0x4e1f;
pDataOutW[2]=type;
pDataOutW[3]=WORD(size*8);
_swab((char*)buf,(char*)&pDataOutW[4],(int)size);

new behaviour :
pDataOutW[0]=0xf872;
pDataOutW[1]=0x4e1f;
pDataOutW[2]=type;
pDataOutW[3]=WORD(size*16);
_swab((char*)buf,(char*)&pDataOutW[10],(int)(size*2-2));

Revision 1460 : the checkbox has been moved to the codec section, but it is not enabled by default

[haruhiko_yamagata]

I tried to playback a dts-wav file (*.wav) with MPC and ffdshow v1381 , but no success ...
Filters: Wave Parser , ffdshow

in ffdshow codecs: DTS in wav is checked (nothing happens)
AC3filter can playback it (if I check PCM at system tab)

Anyone confirm it ? Or do I need something to set?
If it's a bug, can someone tell a working version number?

Thanks
I have the same issue. It's not likely to be a decoder's bug. Something in the registry is broken. I remember system recovery of Windows Xp have worked for me.
Can anybody tell me why?

[haruhiko_yamagata]

Postprocessing doesn't work with FLV4 in new ffdshow builds.
OK, it's easy to add FLV4 (VP6F) to the pp list.
But is pp useful for FLV4? I have only one sample which does not need pp.

[cbemoore]

Ok I am the one who implemented the new SPDIF format, and I didn't figure out why it doesn't work on some receivers.

Hi albain,

I'm not an expert, but could the problem be due to the commented-out AC3 syncword?

The ffdshow code says:

//pDataOutW[4] = 0x0b77; // AC3 syncword (removed because works only for DTS, not for DD)

I've looked at a few other code samples on the web, and they all seem to include the AC3 syncword. Maybe re-adding it would solve the problem?

Chris

[foxyshadis]

OK, it's easy to add FLV4 (VP6F) to the pp list.
But is pp useful for FLV4? I have only one sample which does not need pp.

I've found that inloop is normally on but can be disabled. Hm. Can add a force PP checkbox, maybe, or see if lavc's pp-enabled field is more reliable now. I'll search out some flv4 and see what happens.

Slightly off the topic, might add h.264 deblocking back in, as another optional algorithm to complement the mplayer, nic's, and spp currently. I'm thinking that's what I should have done originally, but what do you think?

[Inventive Software]

FLV4 had inloop? Then why do Youtube videos look so bad? :confused:

And foxy: are you talking about adding an option enabling / disabling the H.264 inloop deblocking filter, as opposed to adding H.264 postprocessing deblocking? The latter isn't good, the former is a good idea. ;)

EDIT: On second thoughts, the method you use probably wouldn't work for most video. I think it's only unique to H.264 that particular inloop deblocker.

[Leak]

FLV4 had inloop? Then why do Youtube videos look so bad? :confused:
IIRC those are mostly FLV1, aka Sorenson Spark aka (almost) H.263.

[albain]

//pDataOutW[4] = 0x0b77; // AC3 syncword (removed because works only for DTS, not for DD)

I've looked at a few other code samples on the web, and they all seem to include the AC3 syncword. Maybe re-adding it would solve the problem?
Chris

I removed it because it made AC3 streams not work (only DTS did).
I had a look on ac3filter code, and it seems to be far more complicated than FFDShow code, and maybe this is the issue : the buffer may need to be scanned for syncword...

[foxyshadis]

FLV4 had inloop? Then why do Youtube videos look so bad? :confused:

And foxy: are you talking about adding an option enabling / disabling the H.264 inloop deblocking filter, as opposed to adding H.264 postprocessing deblocking? The latter isn't good, the former is a good idea. ;)

EDIT: On second thoughts, the method you use probably wouldn't work for most video. I think it's only unique to H.264 that particular inloop deblocker.

Inloop disabling is already available on the codecs page. I mean adding a PP option, but unrelated to whether it's playing h.264 or not, it'd be available for mpeg-4 and other streams too. Same algorithm Donal Graft's Deblock uses, and it works pretty well there, just has a different character than mplayer deblocking.

Some of the early Google Videos were done in VP6, until they switched to Spark everywhere (yuck).

[albain]

One more thing : the new SPDIF behaviour also solves the messup with the audio channels on some multichannel formats.
Besides ac3filter also uses 10 buffers (new behaviour) instead of 4 (old behaviour), so I do believe that the solution is not far

Otherwise I have tried some other methods to fill in the buffer with no success for now

To people where the new behaviour does not work : please tell me if the no sound issue occurs for any audio format (5.1 dolby digital, 5.1 dts), and if you have dev skills you can try to play with the buffer filling (in tffaudiodecoder.cpp / deliversamplespdif). The problem is that I am completely blind on that subject

[haruhiko_yamagata]

@haruhiko_yamagata here is a picture of the bugI can't reproduce. Could you give a try on the downloads from souceforge.net?

[haruhiko_yamagata]

I mean adding a PP option, but unrelated to whether it's playing h.264 or not, it'd be available for mpeg-4 and other streams too.It's a good idea. Could you write the code?

[Shinjite]

Is the latest lavc h264 decoder multi-threaded?
Does it have an impact on P4HT procs?

I have tested clsid's rev 1465, it seems like the H264 decoding still fully utilises 1 thread instead of 2....

[Sharktooth]

the very latest lavc uses multiple threads for decoding h.264 BUT only if the stream was encoded with multiple slices.

http://forum.doom9.org/showthread.php?t=129702

[cc979]

I can't reproduce. Could you give a try on the downloads from souceforge.net?

i checked svn 1443 that is fine, but around svn 1451+ it started - maybe the text is stored different but not sure

[Kado]

Something regarding subtitles... I've been comparing (again) ffdshow renderer and MPC renderer and i've realized that you can get even better subs from ffdshow if you use resize the video before the subs are rendered.
ffdshow places the subs over the video and mpc feeds them to the renderer, thats why (probably) we can't have s subtitle resolution option like mpc has because in ffdshow the subs are dependent on the video resolution.
what's the best settings for resizing a video that can match haali renderer's quality, I'm resizing to 1680pix horizontal.
also, haali resize is done in hardware right (ps2.0)?

Is the resizing done in a different thread from the decoding as well?

[foxyshadis]

Haali uses bicubic, and in hardware, yes. (But it's configurable, whereas ffdshow only has one sharpness setting, 0.6 iirc. Also apparently it uses ps 1.4 now.) MPC's subtitle alpha blending is also in hardware, so it gets a free speed boost ffdshow can't. You'll probably want to use Letterbox 16:10 in subtitles after the resize.

I'll see what I can do for the PP stuff.

[Kado]

thanks for the info. I've noticed that subs have less resolution when using haali if compared to vmr9 and evr from mpc subs renderer.
http://pwp.netcabo.pt/kado/doom9/clip_2.pnghttp://pwp.netcabo.pt/kado/doom9/clip.png

I have found some positioning errors when using ffdshow subs renderer, japs text goes on top of english text, luperry.com in wrong place etc, can be seen at min 1.06
http://img.villagephotos.com/p/2004-6/737150/ffdshow.jpghttp://img.villagephotos.com/p/2004-6/737150/mpc.jpg

Sample here (http://rapidshare.com/files/54220775/sample.mkv.html).

Audio and video have nothing to do with the subs but have the same aspect ratio. subs are from "[LuPerry]_Sumomomo_Momomo_-_01_(1280x720_x264).[F5CDB4ED].mkv"

[haruhiko_yamagata]

thanks for the info. I've noticed that subs have less resolution when using haali if compared to vmr9 and evr from mpc subs renderer.I know MPC has better quality, but you can get better quality from ffdshow. First, check haali's splitter settings (double click the tray icon, Options->Input->Load embedded fonts). Second resize to 1680 horizontally and place subtitle after resize (last is best). Use RGB32 output if CPU can catch up. When ffdshow's subtitle filter get complete, the average CPU would be much faster :sly:.
I'm learning what MPC does for the good quality. It's quite well written, but hard to understand for me.

I have found some positioning errors when using ffdshow subs renderer, japs text goes on top of english text, luperry.com in wrong place etc, can be seen at min 1.06Thank you. I can reproduce. The difference in size is more anoying (hard to fix).

//EDIT The positioning errors are collisions, animation effects and rotation ({\frz}). They are not implemented yet. I have plans to do, but at least after a few months.

[Mc Onyx]

It seems that FFMPEG's E-AC3 a.k.a. Dolby Digital Plus decoder is finished and i'm wondering if it's possible for someone to make a FFDShow build, that has the decoder integrated... Thanks

[haruhiko_yamagata]

Something regarding subtitles... I've been comparing (again) ffdshow renderer and MPC renderer and i've realized that you can get even better subs from ffdshow if you use resize the video before the subs are rendered.
ffdshow places the subs over the video and mpc feeds them to the renderer, thats why (probably) we can't have s subtitle resolution option like mpc has because in ffdshow the subs are dependent on the video resolution.
what's the best settings for resizing a video that can match haali renderer's quality, I'm resizing to 1680pix horizontal.
also, haali resize is done in hardware right (ps2.0)?

Is the resizing done in a different thread from the decoding as well?Yes, ffdshow cannot have subtitle resolution option like mpc. It's impossible. ffdshow does the software alpha blending in the main memory.
Resizing is done in the decoding thread. It uses more than two threads if CPU supports it.
Preparing subtitle should be done in separate thread, but currently it is done in the decoding thread. If you are dropping the frame or experiencing a delay when a new subtitle pops up, queing may help.

[SBeaver]

Hi there I have a question, something I might have touched on before.
Is it possible for 32bit ffdshow to pass on ac3/dts somehow to a 64bit application (like ffdshow 64bit for example).
Since ffdshow is pretty much the only 64bit dshow decoder available except some MS decoders, it's hard to stay away from 32bit media players, since the only one for 64bit is wmp11 which has a lot of limitations.
Any direction to another 64bit media player would appreciated

[Leak]

Any direction to another 64bit media player would appreciated
Since you can't mix 32 and 64bit code in an application this is rather unlikely.

Then again, media playback isn't exactly something that begs for a 64bit app anyway in my book... so why not use a 32bit player on a 64bit OS? All the extra address space and registers isn't going to make a world of a difference...

np: New Order - Everything's Gone Green (Singles (Disc 1))

[SBeaver]

Since you can't mix 32 and 64bit code in an application this is rather unlikely.

Then again, media playback isn't exactly something that begs for a 64bit app anyway in my book... so why not use a 32bit player on a 64bit OS? All the extra address space and registers isn't going to make a world of a difference...

np: New Order - Everything's Gone Green (Singles (Disc 1))

ac3/dts passthrough requires a 64bit app, or some kind of hack.
This goes for all vista x64 systems afaik, and so far 64bit ffdshow is the only thing that works.
Which in turn forces me to use a 64bit player and that is where the trouble starts. This should be a more important problem to solve when more and more move to vista x64.

edit: couldn't the 32bit player + ffdshow split out the audio as usual, and then send that to a 64bit application? There must be some line of communication that can be used

[Romario]

Can someone update x64 builds of ffdshow, please? And add multi-thread decoding to 64-bit ffdshow.

Thanks.

[cca]

ac3/dts passthrough requires a 64bit app, or some kind of hack.
This goes for all vista x64 systems afaik, and so far 64bit ffdshow is the only thing that works.
Which in turn forces me to use a 64bit player and that is where the trouble starts. This should be a more important problem to solve when more and more move to vista x64.

Strange, since I use 32 bit ffdshow along with Zoom Player and Media Player Classic and my S/PDIF Passthrough works perfectly. I do need to check the "Maximize S/PDIF Compatibily" checkbox though. That's on Vista 64 with a Sound Blaster X-Fi Platinum.

[SBeaver]

Strange, since I use 32 bit ffdshow along with Zoom Player and Media Player Classic and my S/PDIF Passthrough works perfectly. I do need to check the "Maximize S/PDIF Compatibily" checkbox though. That's on Vista 64 with a Sound Blaster X-Fi Platinum.

Well ever since vista was released I've been reading about people like you who say everything is ok, maybe creative has some workaround in their drivers but it doesn't work for anyone else.

[clsid]

Also make sure you have the two recent Vista Performance and Compatibility updates installed.

[cca]

Well ever since vista was released I've been reading about people like you who say everything is ok, maybe creative has some workaround in their drivers but it doesn't work for anyone else.

Well, I also have an onboard HD Audio card, for which I have installed the drivers for. I can check it out after work, and tell you if it is working too or not.

[cca]

Well, not sure what is happening after all, but it is something regarding ffdhsow. With the 1471 release I am unable to passthough anything, but AC3Filter does it just fine, and it is NOT a 64 bit filter. Also notable, when ffdhsow is used in conjuction with ReClock, it DOES manage to passthrough. It's up to the devs from this point on.

[albain]

Well, not sure what is happening after all, but it is something regarding ffdhsow. With the 1471 release I am unable to passthough anything, but AC3Filter does it just fine, and it is NOT a 64 bit filter. Also notable, when ffdhsow is used in conjuction with ReClock, it DOES manage to passthrough. It's up to the devs from this point on.

Interesting... this is definitely a buffer sizing issue
All the people who have problems own a 64 bits version of the OS (some datatypes have different size in 32/64bits) ?
It is important to know the answer to this question

Concerning the sound card, I have a creative XFI music and everything works fine on my setup. The problem comes from the HT receiver, and the way it handles the SPDIF (unperfect) encoding.

Otherwise the best way to solve this problem is that someone having these problems would test different sizing parameters with the option "maximize spdif compatibility" unchecked, untill he gets some correct sound.
Of course if AC3filter creator can give us some tips it would be much faster...

[cca]

Well, I can indeed confirm that I use Windows Vista 64bit as my main OS. The problems I described were complete failure of ffdshow to connect to the audio renderer in S/PDIF mode, the receiver never had a problem in my case. Istead, what was happening was that AC3Filter was connecting after ffdshow and was doing S/PDIF decoding downmixing again! If I totally remove AC3Filter, ffdhshow just fails to produce any audio, not able to connect to the audio renderer. If I indeed specify ReClock as the audio renderer, everything works flawlessly. If we can figure this out by testing, I volunteer to help if I can. My receiver supports Dolby Digital EX and DTS 6.1.

[SBeaver]

Just so we are clear on this, I hope all of you actually have a way to tell what signal is going to the reciever.
From all information I have gathered there should be no way to passthrough ac3/dts using 32bit filters in vista x64.
At most you'll get decoded and downmixed to stereo pcm.
There is no need to mention xp x64 since it has no limitations in this area.
I have two soundcards myself, one via envy24HT based and on realtek 850 based.
None of these work on 32bit (realtek last tried with drivers sometime this spring, I assume nothing has changed) but both work when using a 64bit filter like ffdshow x64 builds.
Also please note that trying dts and ac3 in vistas audio settings and playing a test sound proves nothing since it is a 64bit app and should always work anyway.
The only systems I've seed with claimed working passthrough are creative soundcards they are either all false claims or creative has a workaround that they use somehow, but then why wouldn't all other manufacturers do the same thing?

edit: regarding reclock, did you use kernel streaming as the audio renderer? perhaps that bypasses vistas limiting audio stack in some way.

[cca]

Just so we are clear on this, I hope all of you actually have a way to tell what signal is going to the reciever.
From all information I have gathered there should be no way to passthrough ac3/dts using 32bit filters in vista x64.
At most you'll get decoded and downmixed to stereo pcm.
There is no need to mention xp x64 since it has no limitations in this area.
I have two soundcards myself, one via envy24HT based and on realtek 850 based.
None of these work on 32bit (realtek last tried with drivers sometime this spring, I assume nothing has changed) but both work when using a 64bit filter like ffdshow x64 builds.
Also please note that trying dts and ac3 in vistas audio settings and playing a test sound proves nothing since it is a 64bit app and should always work anyway.
The only systems I've seed with claimed working passthrough are creative soundcards they are either all false claims or creative has a workaround that they use somehow, but then why wouldn't all other manufacturers do the same thing?

edit: regarding reclock, did you use kernel streaming as the audio renderer? perhaps that bypasses vistas limiting audio stack in some way.
No, I did not used Kernel Streaming, I have it setup to use WaveOut for S/PDIF in ReClock. Also, AC3Filter can do S/PDIF with both my SoundBlaster X-Fi and my onboard SoundMAX audio device. Beware though: I also have a Logitech Webcam with microphone. I cannot get S/PDIF to work if the that microphone is left enabled in device manager, I have to manualy go and DISABLE totally the mic, which appears as USB Audio device. Sounds strange, but it happens exactly like this.

[SBeaver]

No, I did not used Kernel Streaming, I have it setup to use WaveOut for S/PDIF in ReClock. Also, AC3Filter can do S/PDIF with both my SoundBlaster X-Fi and my onboard SoundMAX audio device. Beware though: I also have a Logitech Webcam with microphone. I cannot get S/PDIF to work if the that microphone is left enabled in device manager, I have to manualy go and DISABLE totally the mic, which appears as USB Audio device. Sounds strange, but it happens exactly like this.

I also have a usb audio device, I will try and remove it and see if it changes anything.
What was the audio you played btw? It's better if you say ac3 or dts since SPDIF really goes without saying since you can't pass anything through without it.

[cca]

I also have a usb audio device, I will try and remove it and see if it changes anything.
What was the audio you played btw? It's better if you say ac3 or dts since SPDIF really goes without saying since you can't pass anything through without it.

It was AC3 5.1 if that helps. The Dolby Digital led lit up immediately in the receiver.

[SBeaver]

It was AC3 5.1 if that helps. The Dolby Digital led lit up immediately in the receiver.


OK I hope that's acurate, my reciever can tell which channels are included and such but I guess if it gets a dolby signal something must be working right for you.

I tried again with the same results.
AC3filter shows the level bars and it says input it 5.1 and output is 5.1 and my reciever is still getting pcm stereo.
Then I try and check the use spdif (passthrough) and then I don't get any audio to the reciever at all.
And in x64 media players everything as working