How to decode 6-track APTX-100 (cinema DTS) with the correct channel levels

[2001CluelessDianogas]

You make a solid argument. But if we assume that an audio processor will treat 5.1 PCM the same way as 5.1 DTS/TrueHD (which would make sense, you wouldn't want your calibration to be off because of some variable like that) then there's still the very convincing matter of the LFE matching blu-rays that more-or-less port the theatrical 5.1 when using the +3/6db rule. The Nolan UHDs are good examples, but I've encountered various others throughout the years. There's also the very subjective argument that I've been listening to tracks that have been editing this way for awhile now and I think they sound right, but take that with a grain of salt.

We could always objectively test weather or not LFE output is the same for bitstream/PCM with an SPL meter. If I can find mine I'll break it out and test.

What is the +3dB or +6dB (or +6dB and +9dB claims) even based on reference to though? Have you been doing manual LFE extraction by low-pass each surround and combining the two channels (using surrounds at given level not -3dB for the LFE extraction)? Or using the auto LFE extraction? Or extracting it from already -3dB lowered surround channels?

I just tested the 3.9 version of the plugin and having it extract the LFE itself as well as auto -3dB each surround.
1. it looks different than manual extraction
2. it comes out too low volume, for pre-1999 JP disc it came out much softer than my manual extraction plus -3dB method. I tried adding +3dB and it was actually still lower volume, +6dB was a touch louder, it seems it's maybe -5dB compared to the "manual method plus a -3dB" so using the auto LFE with version 3.9 I think you'd need to give it a +5dB or for pre-1999 and +8dB or so for 1999+ discs (didn't test whether it comes out this low if you don't select -3dB for surrounds though, perhaps it accidentally just extracts LFE from surrounds after whatever leveling you tell it to give them instead of after)

[2001CluelessDianogas]

You make a solid argument. But if we assume that an audio processor will treat 5.1 PCM the same way as 5.1 DTS/TrueHD (which would make sense, you wouldn't want your calibration to be off because of some variable like that) then there's still the very convincing matter of the LFE matching blu-rays that more-or-less port the theatrical 5.1 when using the +3/6db rule. The Nolan UHDs are good examples, but I've encountered various others throughout the years. There's also the very subjective argument that I've been listening to tracks that have been editing this way for awhile now and I think they sound right, but take that with a grain of salt.

We could always objectively test weather or not LFE output is the same for bitstream/PCM with an SPL meter. If I can find mine I'll break it out and test.

If you follow my instructions for making the LFE you get the same level for LFE for Jurassic Park as the guy in this thread got after making his corrections to the old JP scan's Cinema DTS track where he raised it by +3dB.

To get to the same point manually constructing the LFE I had to do a -3dB to that result to get to the same place.

https://forum.fanres.com/thread-4161.html

The whole you add +3dB or +6dB (or even +6dB and +9dB) talk all depends upon how the LFE was initially created (manually or automatic or by what method of manual) and it's not always clear what the various people were referring to.

[2001CluelessDianogas]

Also if you added +3dB or +6dB to the file LFE itself you'd get insane clipping on Cinema TDS trailer/THX/intro files.
I think that is only if you have the plug-in create LFE itself, which you probably should not do anyway.

[2001CluelessDianogas]

And I'm actually getting same LFE level as Schorman's direct Cinema DTS processor recorded output (after he applied -3dB to the surrounds).

[LucasGodzilla]

So a general question regarding the whole attenuation process.

So because a +6 DB boost in the LFE for mixes like Tenet would clip the peaks, one could conservatively instead lower all other tracks by -6 DB to preserve the LFE as-is (turning the mix from 16-bit to 17.5 bit to account for the -9 DB in the surrounds).

However, in theory, if one were to try encode the mix in a 24-bit DTS-HD MA with the dialnorm set to -25 DB instead of the default -31 DB, would this possibly fix the issue of the mix playing back too low?

Similarly, I wonder if this could be used to handle the unorthodox Mulholland Drive situation, where you do the expected -6 DB attenuation for the period, and then set the dialnorm to -22 DB to compensate for Lynch's instructions.

I've been trying to do a bit of testing with this but I haven't been able to find a way to properly measure how the dialnorm affects the mix, since desktop players like VLC and MPC-HC seem to just bypass it. It's actually rather bizarre where I can see the dialnorm was applied according to the encoder's log but I can't find any way of checking / testing the metadata on the actual resultant DTS-HD.

Does anyone know if this is one of those situations where the metadata only applied to actual hardware setups or is there something else going on that I don't know about regarding the way DTS-HD MA handles dialnorm?

Furthermore, on the subject in general, does anyone have any thoughts on the practicality of this solution? It's certainly at least a thought when the other options are letting it clip or having 32-bit Float PCMs, but I'm not sure if I'm just barking up the wrong tree here.

[borisanddoris]

I’ve been thinking about this lately too. If using Foobar to decode, could you decode to 32-bit floating point and adjust as needed? Would this do a better job? I keep meaning to test it but I just haven’t.

[borisanddoris]

Welp, I just tried a sample of a track that had known clipping in the center when decoding straight to 16-bit. I did 32-bit and it's not there. I then lowered surrounds accordingly by -3dB, boosted LFE by 6dB, and then lowered the entire track 3dB before encoding back to 24-bit LPCM.

The results look nice. My cinema is in a transition so I cannot test but I'd be curious to see how well this works. It would be far easier to run at reference with that -3dB attenuation. My goal is to never over process or color the original track but this seems to look right now. Much more right than before. If this is true, goodness sake, there's a lot of projects I'd have to redo.

[LucasGodzilla]

I’ve been thinking about this lately too. If using Foobar to decode, could you decode to 32-bit floating point and adjust as needed?  Would this do a better job?  I keep meaning to test it but I just haven’t.

I know this was something TomArrow was coining years ago here on the forums...

Idk if this is misunderstanding the discussion but I always thought it would be interesting to do the Cinema DTS decoding into a higher bit depth so that there's more headroom. My reasoning is ... maybe the original track was NOT clipped, but the loss of precision through the band splitting, encoding etc. might have resulted in clipping when everything is put back together.

So basically, the individual bands and whatever would be decoded and then added back together and maybe that's the point where the clipping is introduced perhaps. Now if you decoded into, say, a floating point buffer, for the summing of the bands, then maybe this could be avoided.

I actually asked the developer of the plugin if he could do that but he wasn't willing to do it and it's not open source either so I can't do it myself. Has bummed me for a while ... maybe if someone else were to bum him about it in a way that does not seem coordinated, he might reconsider?

Well my theory (or maybe I should call it a hunch to not overstate it) is that the data inside each band is still not clipped. Basically, the codec, afaik, separates the audio into different bands by frequency before applying further compression. Each band gets its own specific compression ratio and whatnot.
Now, the process would be: separate bands, then encode each band, and write all into CDTS.
So the reverse process would be: decode each band, sum them, and output 16 bit.
Once you have decoded each band and you're summing them up again, there would be nothing stopping you from just allowing for some headroom. Specifically, let's say the normal output is 16 bit. Well, that's a so-called short in programming. You can just use a normal 32-bit integer to write the data into and anything that would have clipped at 16 bit no longer does.
Whether that's actually how it works.... idk. Just a hunch.

I don't think anyone ever did any hard testing though. Foobar does seem to just default to 16-bit on export and it doesn't seem like there's any actual benefit dumping the AUD out in anything higher outside of just general overkill security.

Welp, I just tried a sample of a track that had known clipping in the center when decoding straight to 16-bit.  I did 32-bit and it's not there.  I then lowered surrounds accordingly by -3dB, boosted LFE by 6dB, and then lowered the entire track 3dB before encoding back to 24-bit LPCM. 
The results look nice.  My cinema is in a transition so I cannot test but I'd be curious to see how well this works.  It would be far easier to run at reference with that -3dB attenuation.  My goal is to never over process or color the original track but this seems to look right now.  Much more right than before.  If this is true, goodness sake, there's a lot of projects I'd have to redo.

Reading this though does tempt me to experiment more with decoding formats; maybe there's a way to force a 32-bit output. 

For the sake of science though, I did convert the Tenet R6 AUD to both 16-bit and 64-bit Float to see what'd happen, and both seem to report back with the exact same results in Reaper when doing a dry pass with the original sample rate preserved (resampling to 48000 causes it to peak by +0.2, so I'm guessing it's the r8brain algorithm to blame).



What track decoded with a clip for you out of curiosity?


Also, on the note of the LFE attenuation, does any mix to your knowledge ever clip with a +3 DB boost?

I know Se7en had a moment on R3 when the SWAT team kicks down a door causes the boosted LFE to nearly hit the 0 DB mark, but didn't actually go over the limit. Still, having seen that, it did spook me enough to avoid boosting the channel at all since I didn't want to handcuff myself to a case down the line where +3 DB does end up peaking.



Is this a result of how Foobar2000 decodes the channel or just a coincidental mixing choice where someone in the 90s chose to compress the LFE just below the limit for theatrical playback (which could be notably indicative since skimming down the thread, there were talks of a possible +9 DB boost for pre-1999 mixes)?

[Soundman]

If DTS-6AD decodes with the same volume peak for L,C,R as foobar so why you care about the clicks. I am considering to buy DTS-6AD so i will do some tests!

[stwd4nder2]

Something to note, the DTS-HD MA on the UHD (which seems to be a direct port of the Cinema DTS) also contains clipping on the LFE channel:


Some other titles that are also close ports of the Cinema DTS show the same.