LAV Filters - DirectShow Media Splitter and Decoders

[BloodySword]

LAV Filters 0.40 bug report (splitter in particular):

MKV[H.264+Vorbis] sample (http://www.degeelebosch.nl/reino/H2O - Just Add Water_sample.mkv)
For some mysterious reason the LAV Splitter refuses to connect to FFDShow's Audio Decoder. The LAV Audio Decoder always kicks in. Just try this sample in MONOGRAM GraphStudio and manually add FFDShow's Audio Decoder. The LAV Splitter and FFDShow's Audio Decoder don't seem to agree with each other.

After updating ffDShow, Vorbis won't be decoded anymore because they removed the tremor decoder. I had to enable libavcodec and it worked well for me.

[leewalk]

Then you failed at putting the dts dll in the proper directory, or are using a very old version of the dll and forgot to install the required runtime (which would be automatically installed if you installed TMT. :p)

TMT is installed - latest version. Trying to cover all the bases - still no luck:

C:\>dir "c:\Program Files (x86)\LAV Filters\dts*.*" /s
Volume in drive C is OS
Volume Serial Number is E273-295C

Directory of c:\Program Files (x86)\LAV Filters

10/29/2010 09:31 PM 936,512 dtsdecoderdll.dll
1 File(s) 936,512 bytes

Directory of c:\Program Files (x86)\LAV Filters\x64

10/29/2010 09:31 PM 936,512 dtsdecoderdll.dll
1 File(s) 936,512 bytes

Directory of c:\Program Files (x86)\LAV Filters\x86

10/29/2010 09:31 PM 936,512 dtsdecoderdll.dll
1 File(s) 936,512 bytes

Total Files Listed:
3 File(s) 2,809,536 bytes
0 Dir(s) 53,208,334,336 bytes free

C:\>
Any guess on what might be wrong with my setup?

[Kado]

subtitles with undetermined language will not be displayed by default unless a "und" 3-letter code is added to the subtitle option in the splitter.

Cheers!

[nevcairiel]

subtitles with undetermined language will not be displayed by default unless a "und" 3-letter code is added to the subtitle option in the splitter.



Thats how its supposed to work.

[Reino]

After updating ffDShow, Vorbis won't be decoded anymore because they removed the tremor decoder. I had to enable libavcodec and it worked well for me.I don't see what this has anything to do with Tremor. Libavcodec has been the only Vorbis audio decoder in FFDShow for some time now.
This MKV[H.264+Vorbis] sample is the exception. I have a lot more MKV-files where the LAV Splitter connects to FFDShow's Audio Decoder effortlessly. I just can't figure it out.

[ragg987]

Hi, I am new to LAV and previously used ffdshow. Loving the rules-based subtitle functionality in LAV splitter.

I am tying to use audio decoder as follows:
- if audio is DTS-HD, bitstream
- if audio is DTS, do not bitstream

I wish to use reclock to slow down my PAL DTS DVDs, hence need PCM out with DTS audio.

How can I achieve this? The options in the audio decoder only permit DTS-HD to be bitstream if DTS is also bitstreamed.

Using ATI5750 card with HDMI, Windows 7 32-bit and MPC-HC.

[BloodySword]

I don't see what this has anything to do with Tremor. Libavcodec has been the only Vorbis audio decoder in FFDShow for some time now.
This MKV[H.264+Vorbis] sample is the exception. I have a lot more MKV-files where the LAV Splitter connects to FFDShow's Audio Decoder effortlessly. I just can't figure it out.

Seems to be a problem in the initialization header of the vorbis stream. It is AOTUV, could you check it if you mux standard vorbis?

When I Demux it, the audio file plays normal with LAV Splitter and ffDSHOw, only in MKV it does not play sound at all.

MKV is very buggy I think. I get problems with WAVEPack audio in 32 bit float. It plays as 16 bit signed integer and it sounds slow and distorted.

Just use MP4 with AAC SBR+PS. Sounds not as good as vorbis, but it just works and it is standard.

But I think as soon as LAV Splitter and the Audio and Video filters have the functionality I need wich is in ffDSHOW, I will remove ffDSHOW.

I only need kernel deinterlacer and yadif for interlaced material, audio seems to be ok.

[Qaq]

I am tying to use audio decoder as follows:
- if audio is DTS-HD, bitstream
- if audio is DTS, do not bitstream

How can I achieve this? The options in the audio decoder only permit DTS-HD to be bitstream if DTS is also bitstreamed.
Hmm.. uncheck DTS-HD in LAV at all, use ffdshow for DTS-HD bitstream.

[nevcairiel]

LAV Filters 0.40 bug report (splitter in particular):

MKV[H.264+Vorbis] sample (http://www.degeelebosch.nl/reino/H2O - Just Add Water_sample.mkv)
For some mysterious reason the LAV Splitter refuses to connect to FFDShow's Audio Decoder. The LAV Audio Decoder always kicks in. Just try this sample in MONOGRAM GraphStudio and manually add FFDShow's Audio Decoder. The LAV Splitter and FFDShow's Audio Decoder don't seem to agree with each other.

Fixed


MKV is very buggy I think.

MKV is a container, Containers cannot be "buggy". Its most likely just the muxer that produced something horrible, or the demuxer not reading it properly (or the format just being screwed up in general, like WavPack Hybrid, horrible idea of a format)

MP4 is a even worse container, fwiw.
Take a already horrible container based on QuickTime, and try to add some rather odd ideas an top, and then call it a ISO standard .. voila, MP4

PS:
If you ever want something fixed, i need a sample of a failing file.


How can I achieve this? The options in the audio decoder only permit DTS-HD to be bitstream if DTS is also bitstreamed.

Not with LAV Audio. DTS-HD is DTS, just with some extra data, separating the logic for the two would add quite some complexitiy for some rather obscure use-cases. I mean, if you want to use ReClock, just decode all audio.

[BloodySword]

To use binary XML for a container format is a very bad idea. It's just too complicated to deal with, I hate MKV and I always remux files with H264 and AAC to MP4, because it just works - and it just works on my Samsung Galaxy S, too. I got several problems with MKV, so I dropped it.

I did a draft for a simple but universal container format. If you are interested, write a PM.

What is wrong with WAVEPack Hybrid? It sounds 10000x better than any psycho format like mp3, AAC and vorbis, because it does not touch the frequency spectrum. That's why I wanted to use this.

The WAVEPack decoder in libavcodec is not usable, it is very buggy. Sometimes you can hear "ticks" and sometimes any channel gets muted. The original decoder filter of core codec does not work when demuxed from MKV.

@PS: How can I cut an MKV to a piece of 3 seconds? Does MKVMerge support it?

[nevcairiel]

The name of the format is WavPack, not WAVEPack.
I have no idea how good the decoder in avcodec is, and i honestly don't care because its not a format i ever used. The lossless mode seems to decode fine, though.
In any case, there is an official DirectShow decoder, so just use that. :)

[Kurtnoise]

Hi Nev,

I tested your last commits against the test suite (http://www.matroska.org/downloads/test_w1.html) from matroska.org.

All pass well except test4.mkv :

4. Live stream recording

This file is using the EBML feature that allows Master elements to have no known size. It is used for live streams because they don't know ahead of time the size of the Segment (virtually infinite) and even sometimes the size of the Clusters (no caching on the server side). The first timecode of the file also doesn't start at 0 since it's supposed to be a capture from something continuous. The SegmentInfo also doesn't contain any Duration as it is not know.

The sample comes from the Big Buck Bunny open project. It contains Theora video (1280x720), Vorbis audio, uses only SimpleBlock (matroska DocType v2)

A similar file can be created with mkclean using the "--live" option

I don't know if you want to include this feature or not...

[BloodySword]

The name of the format is WavPack, not WAVEPack.
I have no idea how good the decoder in avcodec is, and i honestly don't care because its not a format i ever used. The lossless mode seems to decode fine, though.
In any case, there is an official DirectShow decoder, so just use that. :)

I want to use it. But only when wrapped in MKV, the decoder seems to output an media type similar to s16l although it is 32 bit IEEE float. So video is half as fast as it should be and the sound makes me dizzy when listening to it: White noise on the left channel and distorted music on the left.

How can I cut any MKV without much effort to provide you a sample file?

[nevcairiel]

All pass well except test4.mkv

I dont think thats a feature that is used by anyone. Also, no MKV splitter i know of can read them, which even reduces the chances that such files exist.

[Reino]

When I Demux it, the audio file plays normal with LAV Splitter and ffDSHOw, only in MKV it does not play sound at all.That's correct, same here.
Thanks for the fix nevcairiel, though I still wonder why it refused to connect. I haven't done anything differently (with x264, nor oggenc2 (aotuv), nor with mkvmerge).
@PS: How can I cut an MKV to a piece of 3 seconds? Does MKVMerge support it?http://forum.doom9.org/showthread.php?p=1050021#post1050021

[Midzuki]

What is wrong with WAVEPack Hybrid? It sounds 10000x better than any psycho format like mp3, AAC and vorbis, because it does not touch the frequency spectrum. That's why I wanted to use this.

*Because* lossy WavPack doesn't use any "psycho-sorcery" :p , it requires more bitrate for sounding acceptable. But IF bitrate and storage space are not a concern, THEN why not use MP1 :devil: or DTS :D

Also:

Why Lossless Audio Codecs generally suck

http://codecs.multimedia.cx/?p=313

The biggest curse in codec design

http://codecs.multimedia.cx/?p=315

[nevcairiel]

Why Lossless Audio Codecs generally suck


Lossless codecs only suck or dont suck if you want to write a decoder for it - otherwise, they are lossless, and there is nothing to "suck". Kostya likes to ramble about how everything is so bad in the codec world - don't mind him. :p
I fail to even see a conclusion in his post, it seems to just go around the fact that there are so many of them.

[rica]

Hi, I am new to LAV and previously used ffdshow. Loving the rules-based subtitle functionality in LAV splitter.

I am tying to use audio decoder as follows:
- if audio is DTS-HD, bitstream
- if audio is DTS, do not bitstream

I wish to use reclock to slow down my PAL DTS DVDs, hence need PCM out with DTS audio.

How can I achieve this? The options in the audio decoder only permit DTS-HD to be bitstream if DTS is also bitstreamed.

Using ATI5750 card with HDMI, Windows 7 32-bit and MPC-HC.

DTS-HD is not an individual format; it always includes dts core and without that core it wouldn't work seperately.

DTS-HD Audio consists of a backwards-compatible 5.1-channel core (44.1 kHz or 48 kHz sampling frequency) and new
coding extensions for improved audio performance and features to support new high definition HD DVD and Blu-ray Disk optical disk formats.
On the other hand DolbY's THD is individual.

Finally, you can not decode DTS core seperately but you can bitstream dts or dts-hd alone. (just untick dts or dtshd options.)

[Kado]

Thats how its supposed to work.

Was loading fine before the 0.38 version that changed the subtitle stuff...

[Midzuki]

Lossless codecs only suck or dont suck if you want to write a decoder for it - otherwise, they are lossless, and there is nothing to "suck".

Demonstration needed :)

And BTW, please don't forget to visit "The Suck Report (http://suckreport.com/)" :D on a weekly basis.

[ragg987]

Hmm.. uncheck DTS-HD in LAV at all, use ffdshow for DTS-HD bitstream.Not sure how to get both ffdshow and LAV audio to do this and switch automatically depending on audio stream.

Not with LAV Audio. DTS-HD is DTS, just with some extra data, separating the logic for the two would add quite some complexitiy for some rather obscure use-cases. I mean, if you want to use ReClock, just decode all audio.I accept that this is not a priority for you, using full decoding of both DTS types for now. Thanks for the programs.

[ragg987]

DTS-HD is not an individual format; it always includes dts core and without that core it wouldn't work seperately.Understand this, however ffdshow has the option to select DTS-HD as bitstream and DTS-HD as PCM. Nevcariel believes this is not a mainstream requirement, which I am OK with.

[rica]

Understand this, however ffdshow has the option to select DTS-HD as bitstream and DTS-HD as PCM. Nevcariel believes this is not a mainstream requirement, which I am OK with.

In my trials ffdshow never decoded core dts of a dtshd stream. BTW you can decode full dtshd with LavAudio which has to use TMT's dtsdecoderdll.dll. But it can never decode core dts by alone like ffdshow can not do either. As for the dts DVD's i will check them and report here.

[BloodySword]

*Because* lossy WavPack doesn't use any "psycho-sorcery" :p , it requires more bitrate for sounding acceptable. But IF bitrate and storage space are not a concern, THEN why not use MP1 :devil: or DTS :D

Also:

Why Lossless Audio Codecs generally suck

http://codecs.multimedia.cx/?p=313

The biggest curse in codec design

http://codecs.multimedia.cx/?p=315

320 kbps MP3 sounds very bad to me, WavPack sounds very good. Also MP1, MP2, MP3 don't support 96000 Hz Samplingrate and even DTS can destroy the frequencys and DTS is highly patented.

For those with damaged hearing (physical or psychologically), MP3 and co might be good, but not for me.

The reason for the problematic packaging and many container formats is the big block size of about one second. The smaller the frame is, the worser is the prediction and the worser the prediction, the more detail is in the differential signal.

For MKV this should not be a big deal but since MKV has a very bad design, there can always be problems.

[Andy o]

We're not bats like you, so we can't hear above 20 kHz anyway (me personally, about 16.5 kHz).

Seriously now, look into learning how to perform ABX tests and you'll save yourself a few headaches (by realizing that mp3 might not be so bad after all).

[BloodySword]

You can not hear above 20 KHz, that's right. But the higher the samplingrate is, the better the frequencys are stored below 20 KHz. See "Nyquist-Sampling-Theorem".

I testet mp3 for myself and I notice it very strongly, that even 320 kbps does even not suit my needs. Sometimes it is like... I'm hearing any mp3, think the quality is good, then it comes to any detailed part in the song and it is totally messed up so I get heavy migraine.

Personally I find soft quant-error noise not as bad as audiopsychoacoustic artefacts and the artefacts resulting of the way how mp3 and aac works.

[Midzuki]

@ BloodySword: Yes, you have very-good points. :thanks:
I had forgotten :o that formats like .mpa, .dca and .ac3 were designed before the "Hi-Def" era :), so, and sadly, their maximum sampling frequency is only 48kHz :( On the other hand, I still think (because I am no actual audiophile) high sampling frequencies (88.2kHz and beyond) sound best :o when compressed losslessly. Differently from DTS-MA, WavPack CAN do without the "hybrid-mode". And again, except for being "bitrate hungry", lossy WavPack is fine. ;)

[nevcairiel]

Please move any off-topic discussions on what audio format prefer to a more appropriate place, though.

[Andy o]

Sorry Nev, I just want to point out that Nyquist-Shannon actually says the opposite, you don't need more than 2x highest frequency to accurately reproduce the wave. Not going to argue about formats anymore :)

[SamuriHL]

What's the command line option for the installer to specify a location?

[fastplayer]

/DIR="full path"

[SamuriHL]

Thanks!

[BloodySword]

A little sample: Imagine a 20 KHz sinus sampled at 40 KHz. It is no sinus anymore, it's a triangle. So you did not understand the Nyquist-Shannon-Theorem at all. It does not say you don't need any more than 2x than the highest freq, it says you need AT LEAST 2 times the sampling to reproduce it. It's like with images. The higher the resolution, the naturallier the image will look.

@nev: How can I cut an MKV so that I can give you a sample of my WavPack problem?

[SamuriHL]

Darn, it still adds the x86/x64 dirs. I think I'll create my own custom install script for what I wanna do. :D

[nevcairiel]

I've been pondering on creating a new installer script, adding better and smarter options, separating 32 and 64 in the process, but I hate working on the damn script. :p

[SamuriHL]

That'd be nice. For now I'm making a custom script that only deals with 32 bit so that I can use it to do an actuall INSTALL into the MC17 private copy dir. Don't ask. :D I'm insane. But to do that I can't have that x86 stuff on it. I do agree, working on the install script isn't enjoyable. :)

[SamuriHL]

Excellent. So I took your ISS, Nev, and modified it so that it only builds and installs the 32 bit version. I set the default path in the new script to the private copy of LAV in MC17. Now I can run a silent uninstall, and then silent install to update my private copy of LAV. This is useless for almost anyone else, but, for my environment it's pretty awesome. I just have to keep my ISS file in sync with yours. I'm not doing version stuff so I couldn't care less about that. But if you add/remove files I have to modify my script. This is pretty sweet. :) Yes, little things that make my life easier amuse me. :D

[Reino]

@nev: How can I cut an MKV so that I can give you a sample of my WavPack problem?You should have a better look at my replies.

[STaRGaZeR]

Found something interesting today. I was looking at getting rid of the splitter/decoder mess needed to properly generate thumbnails in Explorer x64, so I tried removing everything and leaving LAV splitter+video+audio. Everything worked well until it failed to generate the thumbnail in some files. After some investigation it seems related to subtitles. Files with ass subs failed. After installing ffdshow the same files worked, so after some more investigation I think I found why: the video pin and the subtitle pin have to be connected to the same filter. LAV video has no in subtitle pin so subtitles went to who knows where. The DS graphs below are what graphstudio's "Render media file" outputs with each set of options, and probably the same Explorer is using:

- Situation 1, LAV only (FAILS):

LAV splitter (video pin) --> LAV video (video decoder) --> renderer
LAV splitter (ass pin) --> Internal Script Command Renderer

- Situation 2, LAV+ffdshow with LAV as video decoder and ffdshow as subtitle renderer (FAILS):

LAV splitter (video pin) --> LAV video (video decoder) --> ffdshow RAW (video PP) --> renderer
LAV splitter (ass pin) ----------------------------------> ffdshow RAW (video PP)^

- Situation 3, ffdshow as decoder but not accepting subs (FAILS):

LAV splitter (video pin) --> ffdshow (video decoder) --> renderer
LAV splitter (ass pin) --> Internal Script Command Renderer

- Situation 4, ffdshow eating both video and subs (WORKS):

LAV splitter (video pin) --> ffdshow (video decoder) --> renderer
LAV splitter (ass pin) -----> ffdshow (video decoder)^


To tell ffdshow to accept or refuse subtitle connection you have to enable/disable the checkboxes of enabled formats in the subtitle filter. It doesn't matter if the filter is enabled or not.

Do you guys know if this is a known issue and/or fixable in any way?

[nevcairiel]

Its probably not "to the same filter", its more likely the "Internal Script Command Renderer" is doing shenanigans again. Its no good, and i should probably just deny connection to it inside LAV Splitter - it serves no real purpose.

[clsid]

One solution would be to let the splitter expose no subtitle pin when loaded by Explorer.

Here is a FLV sample (http://www.mediafire.com/?404hbc6reir4fhh) that doesn't get a thumbnail. Possibly do to dynamic resizing/cropping?

[STaRGaZeR]

Its probably not "to the same filter", its more likely the "Internal Script Command Renderer" is doing shenanigans again. Its no good, and i should probably just deny connection to it inside LAV Splitter - it serves no real purpose.

Dunno but situation 2 fails too, and it doesn't have that thing in the graph.

[nevcairiel]

Dunno but situation 2 fails too, and it doesn't have that thing in the graph.

Well the ISCR causes issues in other situations as well.

Disabling subtitles when the thumbnail-generator loads it is probably nothing that would hurt either.

[Andy o]

A little sample: Imagine a 20 KHz sinus sampled at 40 KHz. It is no sinus anymore, it's a triangle. So you did not understand the Nyquist-Shannon-Theorem at all. It does not say you don't need any more than 2x than the highest freq, it says you need AT LEAST 2 times the sampling to reproduce it. It's like with images. The higher the resolution, the naturallier the image will look.


Open another thread if you wanna keep discussing this, but there's plenty of info around the net. It doesn't work like that.

[AmshTemp]

Is this reported issue of subs and thumbnailing releated?
http://code.google.com/p/lavfilters/issues/detail?id=68

Anyway I am now using Icaros instead of Windows thumbnail generator, and although it is a little bit unstable it really wokrs most of times... say 98% of the times.

[e-t172]

Sorry for feeding the OT, but I have to do this (http://xkcd.com/386/).

A little sample: Imagine a 20 KHz sinus sampled at 40 KHz. It is no sinus anymore, it's a triangle.

It may look like a triangle to you, if you're drawing straight lines between samples. Fortunately, Digital-to-Analog converters don't work like that. They follow the Whittaker–Shannon interpolation formula (http://en.wikipedia.org/wiki/Whittaker%E2%80%93Shannon_interpolation_formula), which will output a pure 20 kHz sinusoid, not a triangular waveform.

So you did not understand the Nyquist-Shannon-Theorem at all. It does not say you don't need any more than 2x than the highest freq, it says you need AT LEAST 2 times the sampling to reproduce it.

You are completely wrong, and Andy o is right. Let's read the theorem again:

If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart.

Read the bold part again. Protip: in this context, "completely determined" means "lossless".

It's like with images. The higher the resolution, the naturallier the image will look.

Not at all. The human perception of the spatial resolution of an image as observed by the eye has nothing to do with the human perception of the temporal resolution of sound.

[Andy o]

OK sorry nev, I'm starting a thread myself where we can take this.

http://www.avsforum.com/avs-vb/showthread.php?p=21265964#post21265964

I assume most of us are also AVS regulars. Doom9 doesn't seem to have a section for this kind of topic.

[STaRGaZeR]

Well the ISCR causes issues in other situations as well.

Disabling subtitles when the thumbnail-generator loads it is probably nothing that would hurt either.

That sounds good and it'll probably fix it. There is no need for subtitles in a thumbnail after all.

Is this reported issue of subs and thumbnailing releated?
http://code.google.com/p/lavfilters/issues/detail?id=68

I'd say it's the same issue. Don't know about the XP part thou.

[rica]

In my trials ffdshow never decoded core dts of a dtshd stream. BTW you can decode full dtshd with LavAudio which has to use TMT's dtsdecoderdll.dll. But it can never decode core dts by alone like ffdshow can not do either. As for the dts DVD's i will check them and report here.

I checked lav filters with dts DVDs and it works with or without dtsdecoderdll.dll:

http://img221.imageshack.us/img221/1645/lavisworkingfk.png (http://imageshack.us/photo/my-images/221/lavisworkingfk.png/)


http://img849.imageshack.us/img849/7988/lavisworking2.png (http://imageshack.us/photo/my-images/849/lavisworking2.png/)

_ _ __ _

[Pat357]

To use binary XML for a container format is a very bad idea. It's just too complicated to deal with, I hate MKV and I always remux files with H264 and AAC to MP4, because it just works - and it just works on my Samsung Galaxy S, too. I got several problems with MKV, so I dropped it. It's not the MKV container that is bad, but you'll find a lot bad muxes everywhere.
A properly muxed MKV should give no problems at all. Also MKV's can handle almost any content there is. I did a draft for a simple but universal container format.
Hmm.. interesting : will we soon see this container in the wild ?
What content can it / can it not handle ? Any plans maybe for writing a muxer/demuxer for it ? What is wrong with WAVEPack Hybrid? It sounds 10000x better than any psycho format like mp3, AAC and vorbis, because it does not touch the frequency spectrum. That's why I wanted to use this. I agree, hybrid wavepack is a very nice concept.
I still have to test the wavepack DS-filter : I wish it can connect to standard splitters like Lav-splitter, Haali and other MKV/MP4 splitters. It seems to have it's own "splitter" though, what makes me think it will only work with that...
The WAVEPack decoder in libavcodec is not usable, it is very buggy. Sometimes you can hear "ticks" and sometimes any channel gets muted. The original decoder filter of core codec does not work when demuxed from MKV. Are you talking about the DS-filter for wavepack ? (=Wavpackdsdecoder.ax) @PS: How can I cut an MKV to a piece of 3 seconds? Does MKVMerge support it? Yup, see the tab "global" (in the MMG GUI) and look for "enable spliiting". It's also possible to do it from the MKVMerge command line using the proper switches.