LAV Filters - DirectShow Media Splitter and Decoders

[mclingo]

nah, unticked, enable mixing also unticked, just installed pot player, default setting not changes, at least twice the volume, no idea whats going on, defo seems to be LAV filters.

[clsid]

Enable Mixer and set it to Stereo. Set center mix level to 1.0. Do not touch any other settings, leave everything at default.

[mclingo]

no difference at all mate, thought it might be an add issue with my Arctis pro wireless headphones but is louder played through my reciever amp and speakers too.

[mclingo]

ok, so if I enable volume equalization in enhancements in the windows audio properties all the volume comes back, are these supposed to be all disabled?

https://1drv.ms/u/s!AgvFafeelEBior9Oe41B57P7ksa24A?e=DqHMfn


EDIT - that only worked for my amp, did nothing for my headphones, so confused.

EDIT 2 - if I remove the steelseries software so the headphones are running on a basic driver i can get the volume back using the equalization tick enable, anyone any idea whats going on here, surely all enhancments should be disabled?

[filler56789]

ok, so if I enable volume equalization in enhancements in the windows audio properties all the volume comes back, are these supposed to be all disabled?

https://1drv.ms/u/s!AgvFafeelEBior9Oe41B57P7ksa24A?e=DqHMfn


EDIT - that only worked for my amp, did nothing for my headphones, so confused.

EDIT 2 - if I remove the steelseries software so the headphones are running on a basic driver i can get the volume back using the equalization tick enable, anyone any idea whats going on here, surely all enhancments should be disabled?

It looks like all those "enhancements" are actually *bugs* :)

[mclingo]

did a full wipe or drivers,. lav and steelseries software. Much better now but there is still a notable difference in volume in LAV playback vs default potplayer and windows movie and TV player, is this normal?

[el Filou]

If you don't use mixing in LAV, then something else down the chain has to do the downmixing to stereo for your headphones. Or are you using a "surround virtualiser" for headphones in the drivers?
Are you using exclusive mode (this bypasses APOs including those not exposed in Windows Sound Control Panel UI but that are still active anyway, could be useful to test)?
There are so many elements in the chain that can possibly be the culprit.

[clsid]

Potplayer can use LAV if you disable its internal decoder. Test if it gives lower volume then as well.

Windows has separate volume level for each individual application. Click on the Windows volume icon, and check that you didn't accidentally set lower volume for MPC.

In MPC you can choose different audio renderers, which use different audio APIs: DirectSound/WaveOut/Wasapi
MPC Options > Playback > Output

Does it also happen with stereo sources, or just 5.1 stuff?

And if it happens only with 5.1, then which sounds are lower volume? Everything, including explosions, or just the voices, which are mostly coming from the center channel?

[mclingo]

HI, yeah I checked the app volume levels, no joy there all 100%.

cheers guys, I have a lot to look at there, i'll do some more diags and report back when I have a rough idea whats going on.

[Grimsdyke]

I have a few MP4 files on my hardrive which seem to be problematic.
According to MediaInfo v21.03 these files are in 24,520 fps and when opened in MPC-BE + internal splitters + MadVR they play just fine.
But using LAV splitters MadVR claims these files are in 48 fps which means I loose frames on that system during playback.
Does anyone maybe now what could be causing this ?

[huhn]

the file is most likely vfr.

[el Filou]

24 to 48 is quite a range! Are they progressive?
You didn't force deinterlacing or something like that in LAV settings?

[Grimsdyke]

24 to 48 is quite a range! Are they progressive?
You didn't force deinterlacing or something like that in LAV settings?

I think I checked/unchecked all settings that I could find in LAV but the only thing that works is using BE's internal splitter(s).

But are these files just weirdly flagged and BE simply ignores that or should LAV be patched ?
Hope Nev will comment.

[clsid]

As always with this kind of problems, if you want an answer about what is wrong, then you need to provide a sample file.

[DragonQ]

I'm sure this has been asked many times but are there any plans to add a "fallback" hardware decoder option? On my desktop I use D3D11 for any content that isn't H.264 or MPEG2, because it doesn't support interlaced video and those two codecs could be interlaced TV recordings. Software decoding is then used for MPEG2 and H.264, which is fine.

However, my laptop is not fine with the same settings because it heats up a lot with loud fans when using software decoding. I essentially need it to use DXVA2 Native for SD/HD and D3D11 for UHD, which isn't possible currently. Alternatively, DXVA2 Native for interlaced content and D3D11 for progressive content, but I know from experience that it's not always easy to tell the difference, particularly with MKV files. Another alternative would be DXVA2 Native for MPEG2/H.264 and D3D11 for everything else? Currently I have to manually change between the two, which is obviously annoying.

[PCU]

Why does LAV not support Black Magic RAW?

[el Filou]

Doesn't look like it's FFmpeg https://ffmpeg.org/general.html#Supported-File-Formats_002c-Codecs-or-Features

[raymondjpg]

I have tried (unsuccesfully) for many years to get english subtitle selection working well in MPC-BE with LAV filters.

No matter what configuration I try with MPC-BE and LAV Splitter, the first english subtitle in a matroska container is selected, whether it be forced, SDH or normal.

Setting LAV Splitter Subtitle Selection Mode to Advanced, and deselecting default track preference in MPC-BE, then "eng|n" in the subtitle field should select a normal english subtitle track, and ignore any english forced or SDH track. Unfortunately not. If the first subtitle listed is SDH then it selects that. Similarly, if the first subtitle listed is forced then it selects that.

Setting "eng|n eng|!h" in the subtitle does not resolve this. I think that I may gave had some success with suppressing the forced subtitle track using "eng|n eng|!f".

I can find no obvious settings in MPC-BE which might override the subtitle selection mode in LAV Splitter.

Can anyone suggest a reason for this behaviour, or offer a suggestion to overcome it?

Currently my only solution is to extract the normal subtitle track and have MPC-BE configured to prefer external subtititles over embedded subtitles.

Windows 10 Professional 21H1
MPC-BE 1.5.8.6302
madVR 92.17
Either XySubFilter, or Internal subtitle renderer
LAV Splitter or LAV Splitter Source 0.75.1 set as preferred External Filter

[glorp]

Setting LAV Splitter Subtitle Selection Mode to Advanced, and deselecting default track preference in MPC-BE, then "eng|n" in the subtitle field should select a normal english subtitle track, and ignore any english forced or SDH track.

I don't think your advanced mode config string is the correct format. It's needs to be pairs of audio:subtitle language codes separated by spaces or semicolons. The meaning of each pair is: When the audio language is this, then select subs like this. The pairs are evaluated in order to find the first one satisfied and processing stops.

A very basic string for English speaking people who need only forced subs for English dialog, but full subs for any other audio language might be:
eng:eng|f;eng:off;*:eng|d
(the smiley is <colon>o)

This will:
-Select English forced subs when the audio is English, or else
-Turn subs off completely if the audio is English, or else
-Select the English default subs if the audio is anything other than English (since the first two rules handle the English audio cases)

There should be a readme.txt file in the LAV installation directory. It describes this and provides some examples.

[raymondjpg]

I don't think your advanced mode config string is the correct format. It's needs to be pairs of audio:subtitle language codes separated by spaces or semicolons.

You could be right, I thought that eng|n in the subtitle field assumed normal english subtitles for all audio languages.

However, setting the subtitle field to eng:eng|n or *:eng|n makes no difference to the behaviour in MPC-BE. If the first listed english subtitle in a matroska container is SDH or forced, that will still be selected over a normal english track that is further down the list.

[Qotscha]

You could be right, I thought that eng|n in the subtitle field assumed normal english subtitles for all audio languages.

However, setting the subtitle field to eng:eng|n or *:eng|n makes no difference to the behaviour in MPC-BE. If the first listed english subtitle in a matroska container is SDH or forced, that will still be selected over a normal english track that is further down the list.

After testing this a bit, it seems to me that LAV does not recognize visual or hearing impaired flags in Matroska at all. In track listing there are no [visual impaired] or [hearing impaired] texts and, when using advanced subtitle selection, hearing impaired tracks are treated like normal tracks (or default and/or forced if you flag them that way).

With TS files those texts are shown in track listing and subtitle selection works correctly.

[glorp]

mkvmerge added new track header flags for things like hearing/visual impaired, commentary, etc. That was a fairly recent addition (version 54.0, Feb 21). I imagine LAV splitter isn't aware of them yet. IIRC it used to look for "SDH" in the Title tag of a Matroska sub track to determine that it was hearing impaired.

[raymondjpg]

After testing this a bit, it seems to me that LAV does not recognize visual or hearing impaired flags in Matroska at all. In track listing there are no [visual impaired] or [hearing impaired] texts and, when using advanced subtitle selection, hearing impaired tracks are treated like normal tracks (or default and/or forced if you flag them that way).

With TS files those texts are shown in track listing and subtitle selection works correctly.

If I mux the Matroska file into a TS file with tsMuxeR 2.6.12, then all subtitle tracks (listed as PGS) have no track names and have the default flag set to "Yes".

Interestingly playing the TS file in both MPC-HC 1.8.5 (fairly old) with internal LAV filters 0.74.0, and in MPC-BE with external LAV filters 0.75.1, LAV splitter appears to have picked up the first listed English subtitle track (SDH in the matroska container), but displays subtitles from the normal track. This is consistent with your contention that subtitle selection is correct in TS files.

However, manually selecting the English subtitle from further down the list (the normal track in the matroska container) in the LAV splitter subtitle track selection still delivers the normal English subtitles! It's as if the SDH subtitle has disappeared. The SDH track is still there, as it can be selected under the Play|Subtitle Track selection in the player.

All tests were conducted with eng|n in the LAV Splitter subtitle field. Changing that to eng|h or eng:eng|h does not force the SDH subtitles, throwing some doubt on the contention that subtitle selection is correct in TS files.

mkvmerge added new track header flags for things like hearing/visual impaired, commentary, etc. That was a fairly recent addition (version 54.0, Feb 21). I imagine LAV splitter isn't aware of them yet. IIRC it used to look for "SDH" in the Title tag of a Matroska sub track to determine that it was hearing impaired.

Remuxing the matroska file with the latest version of MKVToolNix GUI (v59.0.0), and setting the "Hearing impaired" flag to "Yes" makes no difference to the behaviour. With eng|n or eng:eng|n in the LAV Splitter subtitle field, the first listed English subtitle track (SDH) is still selected in preference over the normal track further down the list.

This is consistent with your contention that LAV filters may not have caught up with recent changes to flags in the matroska container, but it also appears that LAV splitter is not recognising "SDH" in the name tag of a Matroska sub track.

This behaviour is not new. I raised a similar issue about three years ago in the MPC-BE forum.

[nevcairiel]

LAV does not even attempt to determine the type of a track based on its name - such detection is honestly rather unreliable in the best of cases, so I also have no plans to do that. So that not working is perfectly expected.
Matroska SDH flags or such are not read yet, but i'll put it on the list to fix.

[raymondjpg]

LAV does not even attempt to determine the type of a track based on its name - such detection is honestly rather unreliable in the best of cases, so I also have no plans to do that. So that not working is perfectly expected.
Matroska SDH flags or such are not read yet, but i'll put it on the list to fix.

Thanks for that advice.

Regards

[Sunspark]

ok, so if I enable volume equalization in enhancements in the windows audio properties all the volume comes back, are these supposed to be all disabled?


Part of it is the codec, and part of it is the player. If the audio codec is E-AC3 and if I am using MPC-HC I have to turn on Normalise and 400% max amplication with regain volume checked in the player's options. What you found in the Windows properties is the same basic idea. If using exclusive mode, you will have to use the player's options because the Windows setting is for directsound output.

[ryrynz]

NVDec support would be awesome for DVD stuff, the difference between it and other deinterlacers is huge based on my own experience.
In the meantime I'll just switch to CUVID manually any time I play a DVD.

[Sunspark]

Think you're on to something there.

I use the System Default renderer (more info about that below).

It's definitely related to 5.1 vs. 2.0. If I uninstall AC3Filter I can switch between the 5.1 and 2.0 tracks in that sample, and the 5.1 track loads Monogram and the 2.0 track loads LAVCAudio.

With lavcaudio set to "prefer":
AC3 2.0 track: LAVCAudio loads (alone) and sounds good.
AC3 5.1 track: Highest merit AC3 decoder is used. Internal, ffdshow etc. LAVCAudio is not loaded at all.

I'll just include some more info in case it helps (this is all with The Gift sample):

With AC3Filter installed:
LAVFSplitter -> LAVCAudio -> AC3Filter -> Audio Switcher -> Default DirectSound Device - CLSID : {79376820-07D0-11CF-A24D-0020AFD79767}

Without AC3Filter:
LAVFSplitter -> Monogram AAC decoder -> Audio Switcher -> Default DirectSound Device - CLSID : {79376820-07D0-11CF-A24D-0020AFD79767}

Default DirectSound Device should be Speakers on Realtek ALC 885 HDAUDIO. Configured for stereo (tried changing to 5.1, even though I don't have enough speakers, but no change in behavior).

Not using custom mapping in MPC-HC audio switcher.

Tried "DirectSound: AMD HDMI Output" (which I use regularly for my TV), and it also goes straight to Monogram AAC (when AC3Filter isn't installed), like the Realtek DS default.

Tried Speakers (without the DirectSound prefix), and lavcaudio now loads, but no audible result (silence).

The only audiorenderer option in MPC-HC that produces proper sound for me without AC3Filter is "Default WaveOut Device - CLSID : {E30629D1-27E5-11CE-875D-00608CB78066}".

So summarized, I guess LAVCAudio is happy to accept 2.0 tracks, but doeesn't want 5.1 tracks when used with with any of my DirectSound renderers. With non-DS renderers the results are mixed.

If you want me to copy/paste the pin info from a particular place in the chain, with or without AC3Filter installed, let me know.

Apologies for necro-quoting a comment from 10 years ago, but it is still relevant today because I was fiddling with it, so I am going to write about it.

I have the current versions of both MPC-BE and MPC-HC installed and currently both of them have matching LAV filter versions for the purpose of this exercise. One key difference with the audio renderer however is that MPC-BE uses its own mpc audio renderer that is not producing identical sound output to the internal audio renderer in MPC-HC (different CLSIDs) even though both of them are set to exclusive WASAPI mode--as opposed to shared mode.

I felt that the output from MPC-HC was noticeably quieter than MPC-BE and I didn't understand why, since one would assume with WASAPI it would be identical and both were set to 100% volume.

Like Ger, I'm also using a Realtek device and I was working with a 6-channel E-AC3 video that is being downmixed to 2-channel stereo over an analog 3.5mm connection.

One challenge I have, is that I don't have good hearing but I do feel given between two samples to choose from that I can identify which one is has more clarity or is louder than the other.

My current understanding of the 3 main audio outputs Windows uses; DirectSound, WaveOut and WASAPI is that directsound which is the primary sound system will mix & resample anything and everything that it processes. Waveout is also part of the OS but will not be mixed/resampled by Windows afaik, and WASAPI is supposed to be completely untouched by the OS.

In MPC-HC eac3 downmixed voices will be loudest and have decent tonality (more body?) when using Default WaveOut Device.

There is still a slight (I feel) listening difference between it and MPC-BE's audio renderer set for WASAPI exclusive mode but I am not able to quantify what it is but volume wise they are relatively comparable.

So, if you're using MPC-HC downmixing to stereo analog, my opinion is that waveout is best. MPC-BE does not have a waveout renderer. It's either use the wasapi one or directsound.

Probably these results are completely different over HDMI or something else.

For most people likely none of this matters because they're probably using digital connections or receivers, but if someone reading this is using a realtek chip analog 2.1 system I'd love to know if you agree or disagree with my listening observations here.

UPDATE: After clsid wrote below in post #24575 I tried shared mode on the internal audio renderer (wasapi) and everything was loud and clear for my analog output. Comparable to directsound. Directsound is also a shared wasapi method in Windows. I am not using waveout (which is a very old output method) and am set for wasapi shared mode in the internal audio renderer.

[ryrynz]

Why not record the resulting audio from both for a better analysis? Could compare them visually as well as audibly.

[Sunspark]

I don't know how, because I've never done it before.

I'm willing to give it a try. Are there instructions you recommend?

Especially if the results can be depicted visually.

[clsid]

Some people seem to think that exclusive mode magically improves audio quality. It doesn't. It restricts audio output to one application at a time. Which one might consider to actually be a bad thing.

Even when using WASAPI, there can still be resampling taking place in the driver.

Even with DirectSound, resampling is of high enough quality that I doubt you can hear any difference compared to no resampling.

The main difference between WaveOut and DirectSound is how the volume setting in the player behaves. This might be the reason for the difference that you are noticing.
With WaveOut 100% volume in player equals the main output volume level in Windows. With DirectSound 100% in player corresponds to the application volume level (which can be lower than main level). Click on the volume icon in the system tray to see application volume level.

If you want higher volume for voices when downmixing, then that can be achieved by changing the center mix level to 1.0 in LAV audio decoder mixing settings.

[el Filou]

One key difference with the audio renderer however is that MPC-BE uses its own mpc audio renderer that is not producing identical sound output to the internal audio renderer in MPC-HC (different CLSIDs) even though both of them are set to exclusive WASAPI mode--as opposed to shared mode. [...] There is still a slight (I feel) listening difference between it and MPC-BE's audio renderer set for WASAPI exclusive mode but I am not able to quantify what it is but volume wise they are relatively comparable.Do you see information in the respective audio renderers' property dialog on whether they are doing the downmixing themself, or leaving that to the audio driver?
From my own personal experience, there is no guarantee even when using exclusive mode that the audio driver won't touch the sound before getting it to the output. For example, my Creative X-Fi will gladly accept multichannel input in exclusive mode even though it only has stereo outputs, and downmix it in the driver.
Exclusive mode only guarantees the sound won't be touched my Windows' mixer, after that what happens is up to the audio driver.
I don't know how, because I've never done it before.
I'm willing to give it a try. Are there instructions you recommend?
I don't have time to do a drawing, but if your motherboard/soundcard has a "line in" 3.5mm input (= blue colour, not pink), you can connect the "speakers" 3.5 mm output directly to it and then record with an application like Audacity during playback.If you want higher volume for voices when downmixing, then that can be achieved by changing the center mix level to 1.0 in LAV audio decoder mixing settings.Indeed, in general if you want best results I find it's preferable to enable mixing in LAV and set it to the same channel layout as your output, rather than to leave it to the audio driver/renderer.

[SamuriHL]

The only issue with having LAV decode audio is that for things like ATMOS you lose the extension metadata. Apple gets around this limitation by decoding to PCM and then encoding that and the ATMOS metadata into a MAT 2.0 signal that's sent to the audio device. Man I wish we could do that on the PC.

[Sunspark]

UPDATE: I won't keep writing about this because at this point it's about the renderer and not the decoder which the LAV filters are, I just wanted to say one last thing to close this off in case it helps someone else. After clsid wrote in post #24575 above, I then tried shared mode on the internal audio renderer (wasapi) and everything rendered loud and clear on my analog output. Comparable to directsound. Directsound is also a shared wasapi method in Windows. I am not using waveout (which is a very old output method) and am now happy with wasapi shared mode in the internal audio renderer.

[el Filou]

LAV has downmixing options even though it's a decoder not a renderer, so I don't think it's off-topic. The issue here was the downmixing and there are many steps in the chain where it can be done.

[Sunspark]

Good point to make because the implementation thereof is another variable affecting output (the # value used, etc). I am doing the downmixing in LAV. So there are actually 3 possible places on the chain for the downmixing to happen; in LAV, the MPC-BE audio renderer which has an optional mixer (MPC-HC does not have a visible one, just an option to ignore the system mixer which I have checked), and the Windows sound system mixer itself. My hearing is not capable to test these 3 at all beyond general perceptual differences such as loudness and maybe clarity.

Strange that shared mode made such a difference compared to exclusive mode on my system. I am uncertain whether this is because of the audio driver implementation by Realtek or if Windows itself does additional processing even though ignore system mixer is checked. The output from LAV is 32 bit float, but in Windows my system is configured as 24 bit/48000. I have no idea if Windows "knows" that the extra 8 bits are padding or if it literally resamples it down to 24.

[el Filou]

MPC-HC's renderer does automatic downmixing if you check exclusive mode. It then says "Processors: Matrix" in the filter info dialog.

The Windows mixer internally works in 32-bit float too, so there's no conversion from the output of LAV. It then converts to the configured output format for shared mode, but 32 float to 24 int audio is a lossless conversion.

[huhn]

float to int and such is a bit is a bit messy in the audio world.

with int the negative audio can be "louder" with float the positive can be louder. which makes it a bit messy in conversation with clipping and such.

32 bit float or 24 bit int are good enough that is doesn't matter but lossless no.

[LigH]

Assuming normalized values (fixed mantissa), it can be equivalent, just omit the mantissa... but there is no guarantee. And floating point samples may exceed the normal range.

[huhn]

http://blog.bjornroche.com/2009/12/int-float-int-its-jungle-out-there.html

looks like fun

[el Filou]

Yes we're mostly talking video here so I was assuming no clipping or near-clipping.
In standardized video material it's quite rare to have the audio get close to 0 dBFS I think.

[PCU]

LAV Filters doesn't support Bink Video v2!

[LigH]

LAV Filters doesn't support more than libavcodec, the core of FFmpeg. Apparently no programmer implemented it for free for FFmpeg, yet.

[Balling]

Thanks.
Some developers told me that's intel's problem, but a workaround on decoder side can solve this.
Maybe you can investigate further once you got a iGPU. Anyway, it's up to you.


edit: mpc-be r6170 (https://sourceforge.net/p/mpcbe/code/6170/) changed codes for this situation. because Intel automatically tries some transformation when it met hdr videos (VideoTransferMatrix = BT.2020) with evr.

That is wrong. It is PQ transfer that means HDR. 8 bit videos even in avc tagged with PQ are still HDR. BT.2020 spec is only SDR, HDR was defined in BT.2100. With fflay you can get correct bt.2020-ncl decoding (that still does not allow for primaries and tranafer changes (thus no HDR and no WCG), alas). ffplay
-vf scale=in_color_matrix=auto,format=gbrp.

[clsid]

Hi nev, just a FYI regarding your GCC 10.3 build. It has a bug in setjmp/longjmp (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100402) which affects the matroska parser. This is currently the #1 crash in latest MPC-HC build.

Could you perhaps make a 10.4-dev build? Meanwhile I will revert back to 10.2.

[nevcairiel]

I don't do pre-release GCC versions. But I can probably update to 11.2 which also has that fixed.

[filler56789]

I don't do pre-release GCC versions. But I can probably update to 11.2 which also has that fixed.

:cool: :cool: :cool:

:thanks:

https://files.1f0.de/mingw/

[bofex]

Hi Nev!
splitter 0.75 can't open the stream!

rtsp://mediacam:mcam5452@192.168.22.140:8554/axis-media/media.amp?videocodec=h264&streamprofile=feratel_1080_stream

until ver. 0.70 (don't test all other newer releases) working fine with stream analysis duration 3000.
(need to add rtsp/CLID manual in reg)

Best Regards
bofex

[clsid]

I don't do pre-release GCC versions. But I can probably update to 11.2 which also has that fixed.Thanks for the new GCC build.

Would you recommend using the posix threads build? Or does it have any disadvantages compared to the regular build?

[richardpl]

LAV Filters doesn't support more than libavcodec, the core of FFmpeg. Apparently no programmer implemented it for free for FFmpeg, yet.

Big Lies.