Dolby Atmos Theatre System - Page 6

That would be the fastest way. I expect it will take some time to build a level of trust that a rendering algorithm can replace crafting each mix. However, I expect time will prove it is possible, and with the ability to hand tune tweaks along the way if needed, it will quickly become the norm workflow. Mix once, render many.


Yes.


While technically possible, I do not see it happening that way. PLIIz and Neo:X use completely different encoding, so it would not be possible to make one 5.1 or 7.1 track that satisfies both.

The native render of a 3D object-based mix to 5.1 (or 7.1) delivery formats will work about as well as any other 5.1 mix, so it will not degrade the PLIIz or Neo:X result, it just won't help height cues in any particular way either.

One area it will make an improvement is in PLIIx compatibility for 5.1 content. Right now, when 7.1 content is downmixed to 5.1 the rear surround pair is folded into the side surround pair. This erases any ability for a decoder to discern the rear from the side signals.

Ideally the Lsr and Rsr signals would be encoded so that a PLIIx decoder could extract them to the correct rear speakers. That's exactly how MDA rendering works, since it's a natural result of VBAP panning. I presume the same is true for Atmos.

To the extent PLIIz and Neo:X work the same as PLIIx, then at least there's some benefit coming their way.

Just thinking about the tiny pinpoint sounds that audio objects can deliver, how about DTS Atoms?

Sorry I missed your question. I was saying that once audio is described as sounds at positions in space, the way it is mapped to speakers can be decided based on other criteria, such as number of speakers available, the size of the audience or sweet spot, the directional accuracy required, etc. I mentioned "conventional" x-y panning expanded to add z: VBAP, WFS, and Ambisonics (or High Order Ambisonics). Each has its own strengths and weaknesses.

Yes to all, but let me comment on the third one. What is great about the object-based playback (i.e., rendering) is that it works with any speaker configuration file. Just load in the locations (which can be automatic as done in Trinnov), and it renders correctly. Every home could have different speaker coordinates (and MLP target), but the renderer is fine with any. And very easy to support multiple LPs if desired.

Yes indeed. Post #69.

Not really the same thing, though, in my understanding.

First, there is no Trinnov encoder. Instead, it relies on comparing the user-selected "ideal" speaker positions for 2.0, 5.1, or 7.1 systems (from a menu) against the actual speaker positions of the system.

If the actual speakers are in the same locations as in the user selected layout menu, the amount of processing Trinnov applies is minimal. As it should be.

If the actual speakers are in non-standard locations, Trinnov will work its magic to represent the original soundfield.

You may say good, instead of mapping the soundfield relative to a static notion of ITU 5.1, let the source bitstream dynamically move those "speakers" in 3D space, and then Trinnov will make it happen for the listener. That brings us back to "there's no encoder." And if you devised one, I'd wager it would look a lot like an object-based system.

So what we're coming down to then is the question of rendering. Object based sounds can be rendered in different ways, such as panning, Ambisonics (or, more generically, high order spherical harmonics), or wave field synthesis. Each of these techniques can be degraded by a poor choice of speaker positions, and to the extent that a Trinnov-like process can mask those imperfections, it could potentially enhance the final performance of these renderers. But my feeble brain cannot see how it could be used instead of them.

Regardless of what may theoretically be possible with Trinnov technology, I think it is fair to say that as it stands today it cannot just slide into a similar role as the other 3D formats already on the table.

[ETA: I wish I would have waited until I saw Sanjay's post. Would have saved me some head scratching! ]


5.1-->7.1, modest change. 5.1-->40.1, big change. More fun. More room for artistic expression.

I think it would be more accurate to say it leaks an entire channel to every other speaker to build the originally intended soundfield. In this respect it is a lot like Ambisonics, and more particularly high order spherical harmonics. Their AES paper states:
In such systems, the signals come from all the speakers, at particular gains and polarities to reconstruct the soundfield. The higher the order, the larger the sweet spot.


Trinnov can move a "speaker" location, but that is a static positional offset. It requires a lot of calculation to determine these 5th-order filter coefficients that create the effect. If it were confronted with moving that virtual source location on a dynamic basis, it would have to complete those calculations in a span of, say, 5 ms, and do so 200 times per second as the program plays. And not just for 7 sounds (speakers), but for every individual sound object in the program. GXMnow mentioned 7.1 bed channels and 16 objects for the Atmos demo mix. That alone is 3x the number of sources Trinnov supports today. Couple that with the dynamic aspect I already mentioned, and it's probably safe to say this is well beyond the realm of practicality for Trinnov as it is currently designed.

To the extent that imm Sound has a working system based on some form of soundfield reconstruction, that at least proves such things can be done, given efficient computational techniques such as described in their paper: