LAV Filters - DirectShow Media Splitter and Decoders

[nevcairiel]

It doesn't really buffer anything. When it gets an input packet, it throws it into the input buffer, and then trys to decode the buffer. All PCM data is send downstream right away, there is no buffering at that end.
Only explanation could possibly be that the input buffer doesn't fill enough in that time, but that doesn't make any sense, the source filter should keep pushing data as fast as it can.

Considering fps is also using LAV Splitter, it even makes less sense, as it parses the audio frames so always full frames are delivered to the decoder, and no incomplete data (so the input buffer shouldn't be used much)

I also tried to reproduce it with various combinations of audio renderers, but i couldn't get it to break.
I'll think about this. Maybe something comes to mind. But you're the only one to report this problem, too. So, hm!

Edit:
@fps::
One thing i was just thinking about, how is CPU usage while playback? LAV Splitter sets the video delivery thread a higher priority then the audio thread, so if your system is too slow, it might cause the audio decoder to not get any decoding time (decoders typically stay on the same thread as the source filter). But for that to happen you would need to be at 100% usage all the time.

Also, which audio renderer did you use?

[madshi]

Maybe running audio and video pins at the same priority would work better for 100% CPU usage scenarios? I've no idea, I've no experience with writing video+audio splitters yet. But then, that would be a splitter change, so it wouldn't explain why fps has no problems when using a different decoder. <sigh>

[fps]

Ok, it seems to be an issue of the combination of splitter and audio renderer, indeed.
First of all cpu usage is about 33% no matter which combination of splitter/audio renderer/audio decoder.
In all my former log files I used the System Default audio renderer of MPCHC.
Changing it to speaker (Conexant High Def) or Default Wave Out Device reduces the stuttering considerably while using LAV Splitter -> LAV Audio Decoder.
According to the OSD there are no more dropped/delayed frames any more, but the playback isn't really smooth compared to when I use LAV Splitter -> ffdshow audio decoder -> system default audio renderer (or speaker (Conexant High Def), same result).

I think it might be a problem with the clock again, since the clock deviation is much more fluctuating when I use LAV audio decoder.
I made some more logs (http://www.mediafire.com/?76xbsaxit9fdkxs) with the different combinations of splitters + renderers. One thing that's strange is one log file is 4 times the size of the others although I let the player run about the same time. Maybe nothing special, but I thought worth mentioning at least.

[nevcairiel]

Can you try the MPC-HC internal decoder?
It uses approx. the same processing model as my decoder.

[fps]

Do you mean this renderer?
http://t1.pixhost.org/thumbs/1976/6059388_audio.png (http://www.pixhost.org/show/1976/6059388_audio.png)
In combination with LAV Audio decoder this renderer doesn't load on my PC. With ffdshow audio the MPC Audio Renderer gets used as it should.

In the case of LAV Audio Default Direct Sound Device is used instead with this pin info:

Filter : Audio Switcher - CLSID : {18C16B08-6497-420E-AD14-22D21C2CEAB7}

- Connected to:

CLSID: {79376820-07D0-11CF-A24D-0020AFD79767}
Filter: Default DirectSound Device
Pin: Audio Input pin (rendered)

- Connection media type:

Audio: WAVE_FORMAT_EXTENSIBLE 48000Hz 6ch 9216kbps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Audio {73647561-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_IEEE_FLOAT {00000003-0000-0010-8000-00AA00389B71}
formattype: FORMAT_WaveFormatEx {05589F81-C356-11CE-BF01-00AA0055595A}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 24
cbFormat: 40

WAVEFORMATEX:
wFormatTag: 0xfffe
nChannels: 6
nSamplesPerSec: 48000
nAvgBytesPerSec: 1152000
nBlockAlign: 24
wBitsPerSample: 32
cbSize: 22 (extra bytes)

WAVEFORMATEXTENSIBLE:
wValidBitsPerSample: 32
dwChannelMask: 0x0000003f
SubFormat: {00000003-0000-0010-8000-00AA00389B71}

pbFormat:
0000: fe ff 06 00 80 bb 00 00 00 94 11 00 18 00 20 00 þÿ..€»...”.... .
0010: 16 00 20 00 3f 00 00 00 03 00 00 00 00 00 10 00 .. .?...........
0020: 80 00 00 aa 00 38 9b 71 €..ª.8›q

- Enumerated media type 0:

Set as the current media type

[nevcairiel]

The decoder, not the renderer. :) Just enable AC3 in the internal filter list.
The renderer does not work with float audio, which my decoder outputs by default on lossy formats.

[fps]

System Default Audio Renderer + internal AC3 decoder works without any problems.

[nevcairiel]

Ok, thanks. I'll check some things then.

[madshi]

@fps, the "big" log actually contains multiple video starts. It seems you forgot to delete/clear the log file before starting the real logging.

I've compared the LAV Splitter -> LAV Audio Decoder log with the LAV Splitter -> ffdshow Audio Decoder log. Here are the results:

LAV Audio Decoder:

00001197 RefClock->GetTime: 437247650000
00001257 RefClock->GetTime: 437248240000, tick dif: 60, time dif: 59.0000
00001264 RefClock->GetTime: 437248320000, tick dif: 7, time dif: 8.0000
00001277 RefClock->GetTime: 437248410000, tick dif: 13, time dif: 9.0000
00001280 RefClock->GetTime: 437248410000, tick dif: 3, time dif: 0.0000
00001291 RefClock->GetTime: 437248410000, tick dif: 11, time dif: 0.0000
00001375 RefClock->GetTime: 437248997709, tick dif: 84, time dif: 58.7709
00001385 RefClock->GetTime: 437249097709, tick dif: 10, time dif: 10.0000
00001388 RefClock->GetTime: 437249097709, tick dif: 3, time dif: 0.0000
00001403 RefClock->GetTime: 437249281459, tick dif: 15, time dif: 18.3750
00001414 RefClock->GetTime: 437249311459, tick dif: 11, time dif: 3.0000
00001417 RefClock->GetTime: 437249311459, tick dif: 3, time dif: 0.0000
00001434 RefClock->GetTime: 437249601042, tick dif: 17, time dif: 28.9583
00001456 RefClock->GetTime: 437249821042, tick dif: 22, time dif: 22.0000
00001482 RefClock->GetTime: 437249971042, tick dif: 26, time dif: 15.0000
00001528 RefClock->GetTime: 437250532084, tick dif: 46, time dif: 56.1042
00001563 RefClock->GetTime: 437250572084, tick dif: 35, time dif: 4.0000
00001630 RefClock->GetTime: 437251541250, tick dif: 67, time dif: 96.9166
00001660 RefClock->GetTime: 437251851250, tick dif: 30, time dif: 31.0000
00001677 RefClock->GetTime: 437251891250, tick dif: 17, time dif: 4.0000
00001743 RefClock->GetTime: 437252464375, tick dif: 66, time dif: 57.3125
00001793 RefClock->GetTime: 437253070209, tick dif: 50, time dif: 60.5834
00001826 RefClock->GetTime: 437253512500, tick dif: 33, time dif: 44.2291
00001861 RefClock->GetTime: 437253762500, tick dif: 35, time dif: 25.0000
00001910 RefClock->GetTime: 437254358334, tick dif: 49, time dif: 59.5834
00001943 RefClock->GetTime: 437254683959, tick dif: 33, time dif: 32.5524
00001977 RefClock->GetTime: 437255013959, tick dif: 34, time dif: 33.0000
00002026 RefClock->GetTime: 437255517292, tick dif: 49, time dif: 50.3333
00002060 RefClock->GetTime: 437255667292, tick dif: 34, time dif: 15.0000

ffdshow Audio Decoder:

00001219 RefClock->GetTime: 422939980000
00001257 RefClock->GetTime: 422939980000, tick dif: 38, time dif: 0.0000
00001262 RefClock->GetTime: 422939980000, tick dif: 5, time dif: 0.0000
00001288 RefClock->GetTime: 422939980000, tick dif: 26, time dif: 0.0000
00001299 RefClock->GetTime: 422939980000, tick dif: 11, time dif: 0.0000
00001324 RefClock->GetTime: 422939980000, tick dif: 25, time dif: 0.0000
00001327 RefClock->GetTime: 422939980000, tick dif: 3, time dif: 0.0000
00001367 RefClock->GetTime: 422940220625, tick dif: 40, time dif: 24.0625
00001389 RefClock->GetTime: 422940450625, tick dif: 22, time dif: 23.0000
00001394 RefClock->GetTime: 422940500625, tick dif: 5, time dif: 5.0000
00001418 RefClock->GetTime: 422940740625, tick dif: 24, time dif: 24.0000
00001475 RefClock->GetTime: 422941305834, tick dif: 57, time dif: 56.5209
00001518 RefClock->GetTime: 422941733959, tick dif: 43, time dif: 42.8125
00001551 RefClock->GetTime: 422942057292, tick dif: 33, time dif: 32.3333
00001600 RefClock->GetTime: 422942557292, tick dif: 49, time dif: 50.0000
00001635 RefClock->GetTime: 422942903750, tick dif: 35, time dif: 34.6458
00001667 RefClock->GetTime: 422943233750, tick dif: 32, time dif: 33.0000
00001717 RefClock->GetTime: 422943733750, tick dif: 50, time dif: 50.0000
00001750 RefClock->GetTime: 422944053542, tick dif: 33, time dif: 31.9792
00001801 RefClock->GetTime: 422944573542, tick dif: 51, time dif: 52.0000
00001834 RefClock->GetTime: 422944891250, tick dif: 33, time dif: 31.7708
00001884 RefClock->GetTime: 422945391250, tick dif: 50, time dif: 50.0000
00001917 RefClock->GetTime: 422945721250, tick dif: 33, time dif: 33.0000
00001967 RefClock->GetTime: 422946221667, tick dif: 50, time dif: 50.0417
00002017 RefClock->GetTime: 422946731667, tick dif: 50, time dif: 51.0000
00002050 RefClock->GetTime: 422947062292, tick dif: 33, time dif: 33.0625
00002100 RefClock->GetTime: 422947562292, tick dif: 50, time dif: 50.0000
00002133 RefClock->GetTime: 422947894792, tick dif: 33, time dif: 33.2500
00002167 RefClock->GetTime: 422948224792, tick dif: 34, time dif: 33.0000
As you can see, the reference clock behaves very differently. With ffdshow, it sits still until 125ms after playback start. But then the reference clock progresses very near to the system clock. With the LAV Audio Decoder, the reference clock starts moving at once, but it moves in a very weird way, sometimes too fast, sometimes too slow. The jitter is so extremely high that madVR constantly changes its decisions on which video frames to display at which VSyncs.

[fps]

OK, interesting to hear.
Thanks for all the quick input to both of you :).

[nevcairiel]

Wonder if float output somehow confuses the audio renderer.

Try configuring ffdshow to only output Float, and try that.

[fps]

Also works flawless.

[nevcairiel]

Well then, i'm fresh out of ideas. I'll go through the code some.

[Mercury_22]

Fps it's not the only one having "not smooth playback" with LAVAudio
.....

Also when using LAVCAudio the playback it's NOT as smooth as when using MPC-HC's decoders (watch the green line in stats) !

But I'm using EVRCP

LAV Audio (http://img192.imageshack.us/i/lavaudio.png/)

MPC-HC Audio (http://img14.imageshack.us/f/mpchcaudio.png//)

[Ger]

I see the same thing in the EVR-CP graphs. The EVR-CP graph looks messy with LAVAudio, but has a symmetrical sawtooth or flat pattern (depending on how well the frame rate/refresh rate matches I guess) with ffdshow audio.

I only switched back to EVR-CP recently to see if an old bug which forced me to EVR Sync in the past is still present. Now, if I use my old renderer, EVR Sync with the option "Sync video to display (http://www.ostrogothia.com/?page_id=1216)", there is no immediately noticeable difference in the graph between LAVAudio and other decoders.

[pirlouy]

When using LAV 0.17, seeking is quite impossible for mpeg splitter with mpeg1 video (old videos). When using MPC-HC MPEG splitter, all is ok.

[dann23]

@nevcairiel
you said that you're planning to make lav splitter compatible with media foundation. when do you plan to do this?

[VenomousNinja]

When using LAV 0.17, seeking is quite impossible for mpeg splitter with mpeg1 video (old videos). When using MPC-HC MPEG splitter, all is ok.
Me too. I've tried some huge *.vob files - won't seek at all. But the built-in MPC-HC seeks well. My system Win 7 SP1 x64, and MPC-HC x64 build 1.5.1.2958.

[nevcairiel]

@nevcairiel
you said that you're planning to make lav splitter compatible with media foundation. when do you plan to do this?
Media Foundation support has been put on hold until i find someone that is willing to work on some other components in the playback graph. I can provide splitter and audio decoder, but i cannot write a working video decoder, video renderer and proper player as well. Until i see more support from other developers for Media Foundation, i don't think i'll continue in that direction.

When using LAV 0.17, seeking is quite impossible for mpeg splitter with mpeg1 video (old videos). When using MPC-HC MPEG splitter, all is ok.

MPEG Seeking problems are ffmpegs fault, there isn't much i can do about that at this time.

I see the same thing in the EVR-CP graphs. The EVR-CP graph looks messy with LAVAudio, but has a symmetrical sawtooth or flat pattern (depending on how well the frame rate/refresh rate matches I guess) with ffdshow audio.


I could reproduce the issue on my work laptop, and found the issue. Its been fixed for the next version.

The problem was that the duration of audio packets was being calculated wrong, which apparently confused the audio renderer because it would then receive overlapping packages. A video renderer would start dropping frames, the audio renderer just screwed up his timings.. interesting to note that you didn't "hear" the problem.

Thanks for everyones input on the problem!

[dann23]

Media Foundation support has been put on hold until i find someone that is willing to work on some other components in the playback graph. I can provide splitter and audio decoder, but i cannot write a working video decoder, video renderer and proper player as well. Until i see more support from other developers for Media Foundation, i don't think i'll continue in that direction.


I didn't asked for video renderer or player. Just the splitter. WMP12 is good enough as player. I'm sure there are people who will need only need the splitter. I tried with divx but subtitles don't work.

[nevcairiel]

There is no video decoder or renderer with subtitle support for Media Foundation.

[dann23]

There is no video decoder or renderer with subtitle support for Media Foundation.

are you saying that it's not tehnically possbile? if so, how can I insert subtitles when I user wmp12?

[nevcairiel]

You can't, unless someone codes a video decoder that supports them, or something like vsfilter that patches them onto a decoded video frame.

Microsoft does not have support for subtitles in their media foundation components. If you happen to get subtitles in WMP12, its much more likely that it did fallback to DirectShow for some reason.

[dann23]

this sucks in a way but I found this plugin for wmp.
http://sourceforge.net/projects/wmpsub/
So if it'n not very difficult to make lav splitter compatible with media foundation, then I'm willing to test it.
the only problem is hardware acceleration and for which formats it's enabled in wmp. does anyone know this?

[Midzuki]

this sucks in a way but I found this plugin for wmp.
http://sourceforge.net/projects/wmpsub/

http://forum.doom9.org/showthread.php?p=1478847#post1478847

[nevcairiel]

So if it'n not very difficult to make lav splitter compatible with media foundation, then I'm willing to test it.

If it would be so trivial, it would've been done already. :rolleyes:

I am all for Media Foundation, but i cannot use it without proper support. I don't like WMP, and the default EVR Renderer is rather crappy. I even wrote a prototype player some time ago, until i noticed how bad the default EVR is. Until someone at least comes up with a proper renderer for media foundation, i have no personal motivation to work on it anymore, because i just would not use it.

I realize that someone has to start somewhere, but its really a bad situation. You cannot build a player just around the MS tools, because the EVR is seriously bad (the one WMP uses is different then the one 3rd party apps can access), and no-one is going to work on a renderer without a player that could actually use it ... so unless enough people come together for a joint project, i don't see much going on in that direction.

[Virtual_ManPL]

@ nevcairiel - could you please look in bug # 31 - Frame dropping in some certain files (https://github.com/Nevcairiel/LAVFSplitter/issues#issue/31) if you will have some spare time
it's the main problem for me right now, making LAV Splitter unusable in my case, because I see frame dropping in many MKV files

[dann23]

Seems I hit a dead end. subtitles works if they are external files but the ones embedded in the mkv container are useless. It seems I'm stuck with ffdshow and lav splitter unless anyone has a better ideea.

[pirlouy]

MPEG Seeking problems are ffmpegs fault, there isn't much i can do about that at this time.

So how does the MPC-HC splitter ? it does not use ffmpeg ? Or it uses a trick ?

[nevcairiel]

the mpc-hc splitter does not use ffmpeg.

In fact, LAV Splitter is the only DS splitter that does. (known to me)

[Mercury_22]

Your fix for LAV Audio it's working !
Now there is a "smooth" playback :p:thanks:

P.S. Will you consider opening a MPC-HC branch with LAV filters integrated (as an option to MPC-HC's internal)? Cause I think will bring more testers for your filters

EDIT: Small request in LAV Audio -> Properties->Status->Channel Output I think it will make more sens to use this order L, C, R, SL, SR, BL, BR, LFE instead of the current one

[nevcairiel]

P.S. Will you consider opening a MPC-HC branch with LAV filters integrated (as an option to MPC-HC's internal)? Cause I think will bring more testers for your filters
I'll do this at some point, yes. But i don't know when.


EDIT: Small request in LAV Audio -> Properties->Status->Channel Output I think it will make more sens to use this order L, C, R, SL, SR, BL, BR, LFE instead of the current one

The order is the same as the data is send in the PCM packets, and that will not change. Would be overly complicated to remap that anyway :D

[josey_wells]

I just want to say that the LAV Video splitter is fantastic. I was having issue with Blu-ray m2ts files in MPC Home Cinema and Zoom Player with Haili and Gabtest with jumpy video and video/audio sync.

The LAV splitter fixed everything and allowed smooth hi-def playback.

Keep up the good work.

[fastplayer]

Keep up the good work.
I agree! :)
Seeking has never been more fun than it is right now with LAV! :thanks:

[Aleksoid1978]

Hi to ALL.
I have issue with LAV + VC1 + Cyberlink Decoder + MKV.

MKV+VC1+LAV+CyberlinkDecoder:
http://img683.imageshack.us/img683/9513/vc1cyberlinkdecoder.th.png (http://img683.imageshack.us/i/vc1cyberlinkdecoder.png/)

MKV+VC1+LAV+MPCDecoder:
http://img844.imageshack.us/img844/4885/vc1mpcdecoder.th.png (http://img844.imageshack.us/i/vc1mpcdecoder.png/)

[mindbomb]

turn on frame time correction

[leeperry]

Hi, I did some tests w/ the audio decoder...but its performance lacks quite a bit for me. Is it still a very early alpha/beta? I get a lot of random sound cuts(I've got an o/c Q9450)...everything's fine w/ ffdshow.

There's a genius plan in progress to remove all 32fp AC3/DTS decoders from ffdshow, so I would very much like to be able to use your decoder...otherwise I'll be stuck w/ an old ffdshow beta build until the end of days. Besides sansnom05's has managed to unlock 32fp in libavcodec(using madshi's patch) for DTS but for AC3, 2 channels are swapped.

[Aleksoid1978]

turn on frame time correction

But with MPCVideoDecoder don't have this issue.

P.S. AVSplitter from russain play fine in MPC-HC - MKV+VC1+Cyberlink without enable "Frame time correction".
link - http://www.multiupload.com/DG5LP37XCN

[nevcairiel]

I have issue with LAV + VC1 + Cyberlink Decoder + MKV.


The problem is that VC-1 timestamps are so screwed up. I added some code to LAV Splitter to fix those timestamps, so ffdshow, MPCVideoDec and the MS DMO Decoder work properly with it.
Sadly, Cyberlink expects the timestamps "unfixed", and fixes them again.
The next version will have a fix for that, to disable the timestmap fixing when the Cyberlink Decoder is in use.

Hi, I did some tests w/ the audio decoder...but its performance lacks quite a bit for me. Is it still a very early alpha/beta? I get a lot of random sound cuts(I've got an o/c Q9450)...everything's fine w/ ffdshow.
There was a bug in 0.17 (and prior) that potentially caused this. Check back with 0.18 once its out.

I just read up on the arguments on the ffdshow thread, and removing the other third-party libs is a good thing, imho. Why he refuses to accept simple patches to output fp32 from avcodec, i don't understand. The patches are so super simple, give me a list of important codecs, and i can make you patches for them, assuming they decode to float internally anyway.
If he seriously waits for ffmpeg to disable the forced sample conversion to int16 ... good luck. They are working on the audio API currently, but only on the user-frontend, not anything decoding or processing related.

[mindbomb]

so, in the next version, you don't need frame time correction for vc-1 and/or mpeg2 anymore?

[clsid]

There's a genius plan in progress to remove all 32fp AC3/DTS decoders from ffdshow, so I would very much like to be able to use your decoder...otherwise I'll be stuck w/ an old ffdshow beta build until the end of days. Besides sansnom05's has managed to unlock 32fp in libavcodec(using madshi's patch) for DTS but for AC3, 2 channels are swapped.The swapped channels is a bug in madshi's path. There is a new test build in the ffdshow topic for fp32.

[nevcairiel]

so, in the next version, you don't need frame time correction for vc-1 and/or mpeg2 anymore?

At least for VC-1, you should've never needed it when using LAV Splitter, at least when running with ffdshow or MPC Video Dec (or MS DMO Decoder, but that sometimes had some other problems)
The change in the next version is only for Cyberlink (and potentially ArcSoft - untested) decoders.

I didn't even know mpeg-2 had the same issue, never noticed it.


The swapped channels is a bug in madshi's path. There is a new test build in the ffdshow topic for fp32.
It indeed is. I created new patches for this, all available in my ffmpeg repository that i use for LAV Splitter/Audio Decoder, also includes a patch for AAC. Feel free to use them (also up-to-date for nearly latest ffmpeg (2 weeks ago or so, not much changed since then))
http://git.1f0.de/gitweb?p=ffmpeg.git;a=summary

[leeperry]

OK, thanks for the infos! and indeed clsid, your patch passes 32fp for both AC3 & DTS...please commit \o/

I still could use an external decoder for the very reason that some 5.1 AC3 streams are apparently tagged as stereo...here's a sample: http://www.mediafire.com/?mn23gioeoug

This makes my automatic profiles in ffdshow fail...it loads the stereo profile http://forum-images.hardware.fr/icones/ohwell.gif

I also had the problem that many(most?) DS audio decoders first tell a dummy number of channels at the first connection, then disconnect/reconnect and give the right one..that also fubars my ffdshow audio automatic profiles big time. AC3Filter and Gabest's MPC audio decoder say 5.1 even if it's stereo, madFlac never lies...neither does Sonic 4.2(but you can't disable DRC in Sonic).

So if your audio decoder could always provide the right number of channels, this might become my new foolproof 32fp AC3/DTS solution..even for wrongly tagged AC3's http://forum-images.hardware.fr/icones/smilies/love.gif

[nevcairiel]

For the initial connection, the channel count provided by the source/splitter is used, because i just don't have any data packages to confirm if its correct (except if the codec has private data thats being submitted with the media type that contains this information - ac3 doesn't have this, afaik).
In theory, LAV Splitter should always get that right though, because it decodes a few packages to get all relevant information out of the stream.

The Sample you linked up there, for example, is properly being detected as 5.1 by LAV Splitter. (Note, however, that it does not usually load for raw .ac3 files, you can make it load the file in GraphStudio to confirm, though.)
As noticed earlier in this thread, my main goal is to provide a nearly perfect environment if using all of my filters. I just have to rely on some information the splitter gives me, and if the splitter is buggy, well, can't help it.

[madshi]

The swapped channels is a bug in madshi's path. There is a new test build in the ffdshow topic for fp32.
It indeed is. I created new patches for this, all available in my ffmpeg repository that i use for LAV Splitter/Audio Decoder, also includes a patch for AAC.
Sorry for the bug in my patch. I didn't notice it since eac3to remaps all channels, anyway.

And my cheers to clsid and nevcairiel. I'm very glad that you guys care about audio quality and are willing to do something about it... :)

[leeperry]

For the initial connection, the channel count provided by the source/splitter is used [..]

The Sample you linked up there, for example, is properly being detected as 5.1 by LAV Splitter. (Note, however, that it does not usually load for raw .ac3 files, you can make it load the file in GraphStudio to confirm, though.)
Ideally, I'd like to stick to HMS..I would only need a 32fp AC3/DTS decoder that states the number of channels to ffdshow audio(that I've forced) after actually checking. If this kind of wrongly tagged AC3 could come up as 5.1 to ffdshow, this would be fantastic. IIRC, Sonic 4.2(using HMS) tells 5.1 to ffdshow for this very sample.

I just read up on the arguments on the ffdshow thread, and removing the other third-party libs is a good thing, imho.

There's a "-3dB dialog normalization" bug in ffdshow's liba52 implementation(I've provided proofs for this in this thread (https://forum.doom9.org/showthread.php?t=156966)), but better get -3dB 32fp than 16int http://img690.imageshack.us/img690/8138/icecreamt.gif

[nevcairiel]

The whole DirectShow filtergraph is already setup before the decoder sees ANY data to decode. If the splitter sends the data wrong, there is nothing the decoder can do, sorry. (Excluding the cases where it can decode the private data in the media type)

[madshi]

The whole DirectShow filtergraph is already setup before the decoder sees ANY data to decode. If the splitter sends the data wrong, there is nothing the decoder can do, sorry. (Excluding the cases where it can decode the private data in the media type)
The decoder can at any time do a dynamic format change. See here:

http://msdn.microsoft.com/en-us/library/dd388731%28v=vs.85%29.aspx

I know this cause I had to implement support for this in madVR... :p ffdshow supports dynamic format changes, too, AFAIK.

[nevcairiel]

The decoder can at any time do a dynamic format change. See here:

http://msdn.microsoft.com/en-us/library/dd388731%28v=vs.85%29.aspx

I know this cause I had to implement support for this in madVR... :p ffdshow supports dynamic format changes, too, AFAIK.

I support this just fine, the problem he has is with ffdshow profiles, which are only based on the initial connection, so he needs that to be correct.

[madshi]

Ah ok, didn't know that the profiles are based on the initial connection, only.