LAV Filters - DirectShow Media Splitter and Decoders

[nevcairiel]

Originally Posted by mr.duck

I was wondering if it were technically possible when LAVSplitter can't find the length so sets the play length to zero, if it could instead set the play length to some very high value like 6 hours? This would allow MPC to be able to skip forward / back for as much as there is the data available to play.

Yes its possible, but i don't know if its really such a good idea to do this, i imagine it would be a cause for confusion more then anything.

If such zero-padding is such a common case, i could try to add logic for trying to detect it and jumping backwards in the file trying to find the end. With a binary search for the end, it will take a second or two extra during opening (depending on your IO speed), but it would manage to find the real duration. Just not sure its such a common situation that spending the time developing this would be worth it.

[nevcairiel]

LAV Filters 0.51

LAV Splitter
- Increased IO responsiveness on high-latency network connections
- Improved timestamp handling for MPEG-PS (.mpg/.vob/.evo)
- Improved support for H264 streams in SSIF MPEG-TS files
- Added an option to control the priority of audio streams for the hearing/visually impaired
- Added new flags to the advanced subtitle selection
  - "h" for hearing impaired
  - "n" for normal streams (no flags)
  - "!" NOT operator for inverting the meaning of the flags

LAV Video
- Improved DXVA2 Native to software fallback in some situations
- Support for YCgCo to RGB conversion
- Added a new smart Stream Aspect Ratio mode
  - Will only use Stream AR when the container does not provide a realiable AR

LAV Audio
- Improved decoding performance of MP4 ALS and a few other formats
- Fixed playback of COOK with the MPC-HC RM Splitter
- Auto A/V Sync is now automatically disabled on pure-audio files to avoid potential conflicts
- Audio Down-Mixing support

Download: Installer (both x86/x64) -- Zips: 32-bit & 64-bit

I wonder if this has been the longest time between two releases, but oh well, i finally considered it finished enough.

Down-Mixing
One of the big new things in this version is of course the new audio downmixing.
In this version, you get the first set of down-mixing functionality, which should probably cover this function for most people.

To use it, basically just enable it and select the target speaker configuration. You can fine-tune some of the other options, but in general it should work just fine with the defaults.

More control over the full matrix will come in a future version (not that soon).

Everything else
Besides the new audio features, there is the usual list of other fixes, improvements and additions, not everything worth mentioning in the changelog above.

As always, if you find a new bug or regression, please report them in as great a detail as possible, with a sample file if available.

Have fun!

[SeeMoreDigital]

Thanks Nev

[mr.duck]

Originally Posted by nevcairiel

Yes its possible, but i don't know if its really such a good idea to do this, i imagine it would be a cause for confusion more then anything.

You are right of course. I am just feeling a bit cheeky asking you to do this just for me

The holy grail I've been trying to achieve with MPC-HC for years now is for it to be able to handle live content as well as 'static' content (is that the right word?). The problem with MPC, is it cannot load filters that update the play length / seekbar as more data arrives. Such a filter is called a 'push' or a 'pull' source filter. I forget which is which, but MPC is only compatible with 1 of them. I just know that if I can persuade you to modify LAVSplitter to set the length to 6hrs when it doesn't find the end of the video, I could use it to work around this limitation in MPC. If you are willing, you could set the option on by the presence of an empty file with a specific name in the current directory? Or with a registry value you must add/modify manually? To avoid you having to deal with lots of people coming here and say "what's this option do?"

Originally Posted by nevcairiel

If such zero-padding is such a common case, i could try to add logic for trying to detect it and jumping backwards in the file trying to find the end. With a binary search for the end, it will take a second or two extra during opening (depending on your IO speed), but it would manage to find the real duration. Just not sure its such a common situation that spending the time developing this would be worth it.

It would be helpful for sure. I would assume all TV recording software would work in a similar way: allocating a chunk of space for the file (the zeros), then filling it up as the data is received. Allocating another chunk of zeros when the current file size has filled up. But speed is good too. If you were going to implement a search for the end of the video, you could design it to find the end with a fairly low accuracy, like within 10MB for example? You don't have to find the exact end because MPC will keep on playing the video if there is more. You just wouldn't be able to seek up to the very end.

[CiNcH]

0.51 is pretty problematic for me. I am using LAV Video within DVBViewer (DXVA2 native). When switching between H.264 HD channels I often get a green image, sometimes the DVBViewer even crashes.

[nevcairiel]

Originally Posted by CiNcH

0.51 is pretty problematic for me. I am using LAV Video within DVBViewer (DXVA2 native). When switching between H.264 HD channels I often get a green image, sometimes the DVBViewer even crashes.

Seems fine here. Nothing really changed for DXVA either since 0.50.5

[nevcairiel]

Originally Posted by nevcairiel

Seems fine here. Nothing really changed for DXVA either since 0.50.5

If you can reliably reproduce it, can you try with this version and see if anything changed?
http://files.1f0.de/lavf/LAVFilters-0.51-dxvatest.zip

[fastplayer]

Is the 0.71 mixing level in the Mixing tab just rounded up (from 0.707) or is it the actual value? Looking at the source it looks like 0.7071 but I'm not 100% sure...

[nevcairiel]

Originally Posted by fastplayer

Is the 0.71 mixing level in the Mixing tab just rounded up (from 0.707) or is it the actual value? Looking at the source it looks like 0.7071 but I'm not 100% sure...

The slider display rounds to 2 digits

[fastplayer]

Thanks for clarifying and thanks for 0.51!

[nevcairiel]

Originally Posted by CiNcH

0.51 is pretty problematic for me. I am using LAV Video within DVBViewer (DXVA2 native). When switching between H.264 HD channels I often get a green image, sometimes the DVBViewer even crashes.

Because the other version i posted would probably not have done much, here is another one to test:

http://files.1f0.de/lavf/LAVFilters-0.51-dxvatest2.zip

Let me know how this one fares.
If anyone else wants to test this one on DXVA things, you're welcome to it.

[Keiyakusha]

Thanks for the new version! Its perfect!

[6233638]

Originally Posted by nevcairiel

Down-Mixing
One of the big new things in this version is of course the new audio downmixing.
In this version, you get the first set of down-mixing functionality, which should probably cover this function for most people.

To use it, basically just enable it and select the target speaker configuration. You can fine-tune some of the other options, but in general it should work just fine with the defaults.

It's fantastic to have this built into LAV Audio now, but can you explain what the difference between "clipping protection" and "normalize matrix" is?

I would have thought that "normalize matrix" would be the default option, but it seems to be "clipping protection".


And has anyone figured out the solution is for DVD playback Macrovision Errors with madVR and LAV Filters beyond 0.50.1 without modifying system files? It only seems to be affecting some people now, so there must be a way around it.

[nevcairiel]

Originally Posted by 6233638

It's fantastic to have this built into LAV Audio now, but can you explain what the difference between "clipping protection" and "normalize matrix" is?

Normalize Matrix will always normalize the matrix, and some people complained that the resulting audio was not loud enough for their low-powered speakers/headphones (because the volume of all channels is reduced to remove the theoretical possibility of clipping)

Clipping Protection on the other hand will not normalize the matrix, and instead work on the assumption that audio is mastered with so much headroom in the dynamic range that mixing will not overflow it (much). Clipping protection will then analyze the audio and check if it would overflow, and if it does, apply a small normalization factor which reduces the volume a bit so it doesn't overflow.

This has the advantage that you basically use the full dynamic range and don't lower the volume of the audio for no good reason. It'll only lower the volume if required, to avoid clipping.

In my tests, on default settings in LAV, it usually ends up overflowing to 1.2 or something, so the clipping protection reduces the global volume by factor 1.2, so the maximum goes back to 1.0. In contrast, a normalized matrix reduces the volume by 2.6.

[fastplayer]

Can you please add that explanation in a tooltip? I have a "feeling" 6233638's question will come up again and again.

[6233638]

Originally Posted by nevcairiel

Normalize Matrix will always normalize the matrix, and some people complained that the resulting audio was not loud enough for their low-powered speakers/headphones (because the volume of all channels is reduced to remove the theoretical possibility of clipping)

Clipping Protection on the other hand will not normalize the matrix, and instead work on the assumption that audio is mastered with so much headroom in the dynamic range that mixing will not overflow it (much). Clipping protection will then analyze the audio and check if it would overflow, and if it does, apply a small normalization factor which reduces the volume a bit so it doesn't overflow.

This has the advantage that you basically use the full dynamic range and don't lower the volume of the audio for no good reason. It'll only lower the volume if required, to avoid clipping.

In my tests, on default settings in LAV, it usually ends up overflowing to 1.2 or something, so the clipping protection reduces the global volume by factor 1.2, so the maximum goes back to 1.0. In contrast, a normalized matrix reduces the volume by 2.6.

Thanks, so is it adjusting dynamically then, or simply reducing the volume by 1.2?

What happens if both are enabled? (I would assume only normalization takes effect, as it should avoid clipping)

Personally I'd rather take the volume reduction of normalization to avoid clipping and DRC.

Any thoughts on expanding the downmixing with HRTF/Dolby Headphone functions? (I don't know if Dolby Headphone is something that could be implemented in a similar fashion to DTS-HD decoding using an external DLL)

[nevcairiel]

Originally Posted by 6233638

Thanks, so is it adjusting dynamically then, or simply reducing the volume by 1.2?

Its dynamic.

Originally Posted by 6233638

What happens if both are enabled? (I would assume only normalization takes effect, as it should avoid clipping)

Since it would never exceed 1.0, clipping protection would just waste a few cpu cycles, but not touch audio.

Originally Posted by 6233638

Any thoughts on expanding the downmixing with HRTF/Dolby Headphone functions? (I don't know if Dolby Headphone is something that could be implemented in a similar fashion to DTS-HD decoding using an external DLL)

No plans at this time.

[ney2x]

Can you add Enable sound compressor or Audio Boost in LAV Audio (just like in Reclock)? I ask this because I only have a cheap speaker and also I want to ditch Reclock permanently. Anyways, a big thanks for the new version.

[nevcairiel]

I also have no plans for dynamic range compression at this time.

[Keiyakusha]

Originally Posted by 6233638

Any thoughts on expanding the downmixing with HRTF/Dolby Headphone functions? (I don't know if Dolby Headphone is something that could be implemented in a similar fashion to DTS-HD decoding using an external DLL)

We was already talking about Dolby thing not so long ago. It will require implementing wrapper (EDIT: not only wrapper but someone need to figure out its api first.) for 3rd party dll that can't be bundled with LAV filters... Even though I want it too, I don't see it happening in LAV...
There is similar opensource stuff which can be found in ffdshow but to me it doesn't sounds good.