Dolby vs. DTS mastering on same disc

[gorman]

I posted this in The Resistance thread but I guess it's relevant here as well.

https://tech.ebu.ch/docs/tech/tech3324.pdf

Page 25:

For example, consider Dolby Digital (DD) or DTS which have been in the market for more than 10 years: Dolby Digital requires 448 kbit/s and DTS still requires around 1.5 Mbit/s for "Excellent" quality. The newer codecs, such as Dolby Digital Plus or Windows Media provide "Excellent" quality only if operating at 448 kbit/s or above.
And this is the follow-up study: https://tech.ebu.ch/docs/tech/tech3339.pdf

Where, on page 20, you can check table n. 10, that shows how DD at 448kbps still produces good results. Note that this second study concentrated more on multiple transcoding scenarios, if I understand correctly, which are probably likely in a broadcast scenario.

So, basically, it's ok to use DTS 1.5 when available. But a better master in DD will most likely sound better than a worse master in DTS.

 

[jimfisheye]

I have the official Dolby media encoder software.
Dolby TrueHD is lossless. Nulls 100% with the multichannel wav source file.

I don't have the DTS encoder suite. I'm pretty sure it's safe to take their word that DTS-MA is lossless. Whole lotta people would have to be lying in cahoots for that not to be true anyway.

So the formats are lossless and the same source put to each would be bit for bit identical. The thing is that sometimes different masters are put to different formats! I've seen an example before where the lossless version of the program featured a blown out volume war style copy while the lowly lossy core Dolby version was an unmolested copy of the master. The intentional novelty destruction caused much more damage than the loss from lossy.

So formats aside, check out what might be hiding on those discs!

Dolby TrueHD+Atmos is in fact lossy though. Maybe that's what the claim above was aimed at? Not compression lossy... The difference audio in the metadata used to null the object channels back out of the bed isn't a straight phase translation like stereo <-> MS. It's more like a Fourier transformation based "center extraction". The mlp compression is lossless. The object extraction is not. I don't hear anything audible from this in A/B but it doesn't null.

 

[mandrix]

Djabe for sure, and I believe Rob Reed have done 24/96 DTS on DVD's and it sounds pretty darn good for "lossy" to my ears.
But then, Tim Bowness newest album in DD ain't too shabby sound wise. Of course I believe Wilson did the mixing, so there's that.

 

[Owen Smith]

Djabe for sure, and I believe Rob Reed have done 24/96 DTS on DVD's and it sounds pretty darn good for "lossy" to my ears.
But then, Tim Bowness newest album in DD ain't too shabby sound wise. Of course I believe Wilson did the mixing, so there's that.

DTS 24/96 is dumb in my view. DTS is lossy and has a fixed bit rate, so why waste some of those bits encoding things that only dogs and bats can hear? Better in my view to use those bits to be less lossy encoding the frequencies people can actually hear.

 

[jimfisheye]

Shades of grey perhaps but I'll take "dumb, but works" over "dumb and doesn't work at all"! DTS2496 is transparent to me. A couple opportunities to compare it to fully lossless copy of the same program had it null down to -90db. That's a hair from lossless with just some quantization noise at -100db.

The "dumb" part is the almost fully botched implementation that resulted in many stand alone disc players, media players, and ripping apps only grabbing the core dts. The corrupted sound from that was beyond the usual lossy stuff and got some pretty bad reviews. Not just 'lossy' when that happened but very audibly corrupted sounding.

There was (and still is) a software solution though so intact audio delivery is possible with caveats.

 

[mandrix]

DTS 24/96 is dumb in my view. DTS is lossy and has a fixed bit rate, so why waste some of those bits encoding things that only dogs and bats can hear? Better in my view to use those bits to be less lossy encoding the frequencies people can actually hear.

Woof!

 

[mandrix]

I'm sure due to licensing fees, production costs, etc, is why some artists turn out DVD as opposed to BD. Pretty sure, but don't know for fact, that someone in the Djabe crew does all the mixing and wouldn't surprise me if they authored the DVD's, thereby cutting costs. Of course they turned out a fair few DVD-A as well.

DVDV or ADVD being the case I'd rather have 24/96 DTS than DD normally.

 

[Zacabeb]

Dolby TrueHD+Atmos is in fact lossy though. Maybe that's what the claim above was aimed at? Not compression lossy... The difference audio in the metadata used to null the object channels back out of the bed isn't a straight phase translation like stereo <-> MS. It's more like a Fourier transformation based "center extraction". The mlp compression is lossless. The object extraction is not. I don't hear anything audible from this in A/B but it doesn't null.

Is it still lossy beyond the Spatial Coding step? That is, once the Dolby Atmos Renderer or RMU has reduced the mix from the original beds and objects to the 11 or 15 object clusters + LFE, is there any additional loss after delivered through the MLP stream?

My belief has been that the fourth substream contained pretty much raw MLP compressed audio difference between the object clusters and the 7.1 render from the other substreams and only E-AC-3+JOC/AC-4+JOC had parametrically extracted object clusters.

 

[jimfisheye]

I haven't studied all the tech papers and I'm pretty sure their exact process is proprietary anyway. Just observing what the software is doing and the resulting files.

The file size of the metadata delivering the object data is too small to be the full object channels. Even compressed to mlp. Dolby claims to fold the object audio into the bed tracks. We see rips from not decoded TrueHD+Atmos result in 8 channel 7.1 files that magically include the object channels folded in. That seems to confirm the speculation on that. (Or the object audio would not be there!)

My speculation is they created difference files for each object channel to null the object audio back out of the bed tracks. Probably Fourier transform based work like the center extraction methods. There's no direct translation like stereo <> mid/side. MLP encoded difference tracks would follow with the size of the metadata.

MLP compression is lossless. Extracting the object channels is not.

The lossy 'JOC' version throws the baby out with the bathwater by hitting everything with lossy compression instead of MLP. The ability to still have a transparent extraction with Fourier transform tricks collapses with the lossy bed and difference audio. Hence streaming Atmos being as mutilated as it sounds.

Just observing. I could just stop at: TrueHD+Atmos does not null with the source 7.1.4 wav file. That seems cavalier to just state that way without more comment though. Every one I've A/B'd on my system sounds identical to the ear but very much does not null.

The same thing happens when playing with stereo center extraction tools. You can mix the outputs back together and it does sound just like the original stereo file. But it sure doesn't null! Not even a little either. Whereas sample rate conversions, for example will null down to -90 or -100db. Just a difference in the lowest noise.

Again, just observing. And I can't help but also observe that the wavpack format suits 12 or 16 channel mixes very well and fully losslessly with no proprietary decoder. Seems like a better choice. Except that wouldn't force anyone to have to buy a new AVR so... TrueHD+Atmos!

Decoding and speaker management would be separate steps. Decode back to 7.1.4 audio. Now speaker manage it to the array available (if it's different than 7.1.4).