Originally Posted by GXMnow
Okay, I am back, had a few long work days.
I also checked with some people at Dolby to make sure I can say these things and was given the ok.
If anyone on here saw "The Incredibles" clip in Vegas, take a guess how many objects were in use?? I will give this at the end of my posts so you have to go down to see it.
The "BED" channels do serve a few purposes. The system config defines what speakers make up each bed. The installer picks which output channel or group of channels are to be used to form each bed. The setup I dealt with, we defined the normal left, center, right, left side, right side, left back, right back, left ceiling, right ceiling, and also defined a front wide left and right. In a different room with an odd number of speakers across the back wall, they did define the center rear, but no contant has called for that as a defined bed. Since the pan off of the screen does work better with side speakers going up much closer to the screen, we actually do leave out the front few speakers from the side beds to better create the normal side surround field. An interesting extra thing is that once the beds are defined, we actually do run an additional EQ pass on any bed that uses more than one speaker. This way, when the group is driven as a single channel, it is the proper frequency response and level. When you group speakers, the way they will add varies depending on the spacing and boudries. This takes care of the issue. Once tuned up a -20 dbfs signal to a single speaker will produce proper response at 85 dbc, and a bed using 8 speakers will also produce the same response at 85 dbc. I am not going to argue response curves at this time, but all channels are using the current X-Curve response. The main LFE stillhas the same +10 in band gain, but any additional subs are only used to fill in for bass management, so will not have the additional gain. Bass managing into the main sub does take this into account and matches the level of the channel it is extending. The current software will only use real speakers to make beds. You can chose any combination of speakers, and this will create a phantom image between speakers, but there is no special processing to try and make a bed where there isn't a speaker. I can not say for sure how a signal sent to a ceiling bed is handled if there are no ceiling speakers. I know it will use the wall speakers, but I do not know if the bed has to be defined with what is there, or if it uses the known positions to create it's own. I will try and get this info next week, but so far all the rooms I have been in had real ceiling speakers so this question did not come up.
As I said before, the software is still going through changes, so a few extra limits have been set for the early tests. All the demo material so far is using a 9.1 bed. 7.1 and the left and right ceiling arrays.
About using nothing but objects...
I will use dialog as an example. The renerer is very accurate, and all the signals and setup info is stored digitally so there should never be any drift, BUT... If you put dialog at a location at the center of the front wall, the renderer will try to locate it right inthe actual center speaker, but I feel this is the same as a Pro Logic center channel as far as accuracy. If there is any error in the setup and rendering, it might have slight crosstalk into Le/Re, or even L, R depending on how close they are or if they might be a bit off center or if the position is entered with less than perfect precision. Using a center channel bed makes the sound always use just the real center speaker, no chance for any crosstalk.
Effects, music, or even a voice moving across the screen will be fading from speaker to speaker anyways, so a tiny bit of crosstalk or even being a phantom image between 2 speakers is not an issue, but for certain things like the dialog, this could be a real issue.
Also, if you just want a nice ambient sound or music spread through the room, the bed array surounds work very well. To get the same spread effect with just objects, you might need a few objects down the length of the desired array. There are many sounds that just work very well in an old style speaker array. But there is nothing forcing anyone to use them, an entire mix can be done with only objects if the mixer wants to do it. Out of 128 objects, we have been setting asside 10 (9.1) for these beds, so you still have 118 dynamic objects, and the production software to come soon will allow defining just a 5.1 bed, leaving 122 dynamic sound objects.
All of the current demos are also using no packing or compression at all. So if an object is used at all, it takes up a full 24/48K PCM track on the hard drive. This will change very soon, at the very least will be zero removal. A track will only be on the drive when it is in use. So when just using beds, just 9.1 will be streaming, or about 12,000,000 bits per sec. When 120 objects are active though, we still jump up to a huge data rate. It can exceed 140,000,000 bits per sec. Everyone on the production side wants lossless so that is the plan. Lossless packing can achieve about 3 to 1 compression on normal audio. Dolby True HD is in this area and I would assume a very similar packing technique will be used, but this will likely be optional, and maybe only invoked when high object counts are in use. Getting the data from the DCI server into the renderer is the possible bottle neck. The Dolby server has Gigabit ethernet for this purpose and should not have any issues even cranking the full 128 uncompressed objects. The DCI spec does already have provisions in it for "aux data tracks" which is what the ATMOS stream will be. So Dolby is not violating the written spec to put this data file along side the existing package files. Dolby will make available any info needed for other server makers to be sure they will be able to stream this data out the ethernet port. It is up to them if they will need to update firmware, or if they will charge for any such upgrade etc.
So, the incredibles clip, with all those sounds flying all over the place, uses only 16 total object tracks. Each track is being used over and over with new sounds every second or so. One second an objec might be the sound of a punch that misses, whizzing by you, but then that track is now the debris bouncing off the mountain as the ship crashes. Then it is the leaves rustling as Dash slides down the vine. Several thousands source elements may be used, but it is rare to need even 50 active objects for any instant. Of course, this clip is using the beds very effectively. Using less effects in the beds will obviously increase the object count, but most feel the currently impossed 128 source tracks is not a problem.
SRS MDA is claiming they are totally unlimited and therefore better than ATMOS as they could render millions of sources to thousands of speaker feeds. This is just plain silly. The 128 source tracks is to be sure there will not be an issue storing and streaming the data in the current server technology. This number could increase with more capable servers and bigger hard drives to support the data needs. More output channels could also be added with a more powerful renderer. Dolby suggests the speakers be 30 degrees apart or less to get a smooth pan. If someone builds a huge theatre and feels they need more than 64, I am sure a renderer can be put together. How many HUGE theatres are left??
As for a home version, yes they are talking about it, and this is a perfect place to have different price points. An AVR with a 10 channel amp and rendering capability would obviously be cheaper than an AV Pre Pro with 32 outputs and rendering ability and separate amps. It is very scalable. Getting the ATMOS data onto a Blu Ray is another discussion.