Cavernize free EAC3 Atmos decoder

[tebasuna51]

Like was mentioned by Balling here there are a free EAC3 Atmos decoder https://github.com/VoidXH/Cavern

I open this new thread to explain some test about it.

I used:

"Cavernize for FFmpeg.exe" input.eac3 -conv output.wav
...
 A 5.1.2 output (L,R,C,LFE,SL,SR,TL,TR) will be used...

to obtain a default correct output than can recoded to eac3 with Audition and played fine with my AVR.

Also can configure my speakers positions:

"Cavernize for FFmpeg.exe" input.eac3 -conv output.wav -count 8 -override 0 0 -45 false -override 1 0 45 false -override 2 0 0 false -override 3 15 15 true -override 4 0 -120 false -override 5 0 120 false -override 6 -45 -45 false -override 7 -45 45 false

Obtaining a different output.

But I can't confirm the decode quality, by the moment I detected a issue

[Balling]

After latest commits it now decodes correctly every time, not different and it also now does not apply DRC (so sound is some little louder) and also it decodes to 24 bit, even ADM. 17 objects are also never reported.

Here I compiled GUI latest: https://github.com/VoidXH/Cavern/fil....0-windows.zip

[tebasuna51]

The latest GUI now does not apply DRC but there are a problem with large files and the 24 bits:

ffmpeg version n5.1-1-g6d83c5f876-20220729 Copyright (c) 2000-2022 the FFmpeg developers
...
Input #0, eac3, from 'C:\tmp\Dragon4.ac3':
  Duration: 01:02:30.02, start: 0.000000, bitrate: 768 kb/s
  Stream #0:0: Audio: eac3, 48000 Hz, 5.1(side), fltp, 768 kb/s
Stream mapping:
  Stream #0:0 -> #0:0 (eac3 (native) -> pcm_s24le (native))
[segment @ 000002513a33aac0] Opening 'C:\tmp\Dragon40.wav' for writing
Output #0, segment, to 'C:\tmp\Dragon4%d.wav':
  Metadata:
    encoder         : Lavf59.27.100
  Stream #0:0: Audio: pcm_s24le, 48000 Hz, 5.1(side), s32, 6912 kb/s
    Metadata:
      encoder         : Lavc59.37.100 pcm_s24le
[segment @ 000002513a33aac0] Opening 'C:\tmp\Dragon41.wav' for writing
[segment @ 000002513a33aac0] Opening 'C:\tmp\Dragon42.wav' for writing

- Decode in segments of 30 minutes and create 3 files Dragon40.wav, 41 and 42.
The 3 files are 24 bits and MaskChannels: 1551 (FL FR FC LF SL SR)
But automatically the 3 files are joined in a Dragon4.wav (Duration: 01:02:30.02) but now at only 16 bits and simple header without channelmask.
The output file must be a .w64 or .rf64 to avoid the size limit of .wav files and preserve the 24 bits precision.

- And I ask for 5.1.2 (see image), for what only 6 channels?
Nothing to TFL-TFR channels?

While eac3 encoders do not read the channelmask it is not important.
For 5.1.2 the channel order seems correct when decode a EAC3 7.1 but for 7.1 we need remap the channels SL-SR <-> BL-BR

[Balling]

ADM is in fact RF64, i.e. w64. if other reconstructed from segments file is not RF64, report that...

[tebasuna51]

Originally Posted by tebasuna51

- Decode in segments of 30 minutes and create 3 files Dragon40.wav, 41 and 42.
The 3 files are 24 bits and MaskChannels: 1551 (FL FR FC LF SL SR)

It is OK, these files are standard decode without apply Atmos metadata.

But automatically the 3 files are joined in a Dragon4.wav (Duration: 01:02:30.02) but now at only 16 bits and simple header without channelmask.
The output file must be a .w64 or .rf64 to avoid the size limit of .wav files and preserve the 24 bits precision.

Not a simple join, the Atmos metadata are used to create the desired output channels. I want 5.1.2 and create 8 channels using rf64 header with big files (Star Wars I: 5.84 GB) but without channelmask and only 16 bits.

BTW recode Atmos to eac3 5.1.2 can be for players without Atmos support or to reduce size, maybe until 448 Kb/s, and the 16 bits can be aceptable.

The channelmask is not important for eac3 encoders because seems is ignored. For me, with the render target 5.1.2 front, I can recode to (L R C LFE Ls Rs Vhl Vhr) without remap.

For me the first feature request by the moment is decode TrueHD.

The ADM format is useless for me.

[Balling]

So I further tested Justice League sample with CavernGUI. It "reports" that it has actually present dynamic objects: 13, which means there is movement. But that means it does not matter whether those are tagged Tfl/Tfr, cause those are extracted from those channels and go into strange places

[Balling]

New caverngui version. With reworked renderer. more correct, tested with Dolby test tones being now all the same amplitude: https://github.com/VoidXH/Cavern/fil....0-windows.zip

VS 2022 17.3.5.

[Balling]

New CavernGUI version that gets ADM master object Atmos file correctly https://github.com/VoidXH/Cavern/fil....0-windows.zip

Works with Dolby Media Encoder.

[tormento]

Originally Posted by Balling

New CavernGUI

How can we get it from GitHub without the direct link?

What is the path where it's stored?

[tebasuna51]

Thanks for the news Balling, but remember the forum rule:

8) No cross posting. Post your message once, to the appropriate forum and nowhere else or it will be locked or deleted without warning.

Here is ok, the other 2 post deleted.

[Balling]

Originally Posted by tormento

How can we get it from GitHub without the direct link?

What is the path where it's stored?

https://github.com/VoidXH/Cavern/iss...ent-1236321425

[tormento]

It's a real pity it can't deal with THD.

[spoRv]

Originally Posted by tormento

It's a real pity it can't deal with THD.

Not yet - non ancora...

[Balling]

Originally Posted by spoRv

Not yet - non ancora...

There is no spec for Atmos in THD. So... It is closed standard. Atmos in EAC3 is very open though.

[Balling]

https://github.com/VoidXH/Cavern/fil....0-windows.zip

Final version, now ADMs with some static objects also work. And with static objects not in standard places.

[KSSW]

I hope metadata can be exported

[Balling]

Originally Posted by KSSW

I hope metadata can be exported

There is an option to write XML metadata separtely when you save ADM.

[johnnyquid]

My TV is in the corner of the room. My room and speaker configuration is therefore non-standard and very asymmetrical. The left and right side speakers are different distance and angles to the listening position. The surround and height speakers are also asymmetrical. I have the relative X, Y, and Z coordinates of the speakers, TV center, and listening position.

Do the inputs for W/Y and H/X make any difference or does QuickEQ make all needed adjustments without using the W/Y and H/X inputs?

Where can I find what W/Y and H/Y represent geometrically in 3D space (i.e. how to calculate using only X,Y,Z coordinates)?

Answers from Cavern author:
QuickEQ does not use the W/Y and H/X inputs. (QuickEQ must be adjusting for volume and frequency response only).
W/Y is the angle off the center line in the horizontal plane.
H/X is the angle off the horizontal plane.
This is what I expected but it is good to have confirmed.

[KSSW]

Whether Cavern 1.6 is 1:1 decoding Dolby Atmos

[Balling]

Latest fixes fixed smaller size of render WAV compared to ADM, fixed render for EAC3 without AC3 core substream, fixed crazy problem with files with dedicated Beds, more than 1 that is; fixed problems with opening ADM files in Davinci/Nuendo/Dolby Media Encoder before Windows 11 22H2, but iXML is now much bigger...

We also discovered bugs in rendered 7.1.4 WAV extensible format in Adobe Audition and in Media Encoder. Hilarious.

There are still problems for older than 22H2 OS and problems with gain.