[Uni-verse] Audio demo situation?

[Samuel Siltanen]

Dear all,

Here is a description of the situation from which we are likely to begin 
on Monday morning, unless there are volunteers to work over the weekend
(I know some of us are).

1) Have two alternatives for the visual renderer: OSG and Quelsolaar. Both 
are running. OSG has been tested with the test geometries and they seem to 
be working as the materials used in them are simple. Also the listener 
position is updated in such a format that UVAS understands it.

Quelsolaar has not been tested as extensively since we had problems with 
the computer hardware. Also cooperation with UVAS is not clear since the 
description of the "listener" object cannot be found in the 
official auxilary Verse specification, which makes it very unprobable that 
it is implemented in Quelsolaar. Of course, the reducer could make the 
conversion from the "camera" to "listener" if needed.

There is no visual counterpart to the sound sources in either of the 
renderers. That is why the sound seems to come from an invisible source in 
mid air. This is something we did not think about until the whole chain 
was under testing. Adding real geometry around the sound source does not 
work since it blocks the sound inside it. 

2) The 3D Studio Max plugin to Verse seems to handle the geometry well and 
sound sources can be added and modified in it. However, electing sound 
sources in Max is difficult without using the menus. This is good enough 
for the demo. 

I guess some of the more experienced Max users could teach how to modify 
the geometry without breaking the model. This is essential if we are to 
show changing geometry in the demo. Could this be automated by using the 
animation properties? Otherwise someone's hand will be aching after moving 
a wall back and forth for three days.

3) The reducer handles the dynamic geometry as well as sound source and 
listener data. Basically, it forwards any data from the main Verse server 
to the reducer server, only reducing the geometry in the process. The 
reduction is as real-time as the functions of the other Verse tools.

There are limits to the amount of reduction. Due to the real-time 
requirements the more advanced techniques like volumetric decomposition 
and coplanar set merging are not feasable. However, the most recent demo 
scene with about 5500 polygons was reduced into 590 polygons while  
preserving its main features.

4) Implementing UVAS has probably been the most demanding part in the 
acoustic simulation system. It has been tested with the previous test 
scene in which it performed well, even with multiple sound sources. The 
new scene is significantly more complex and, based on the initial tests, 
it can be said that its performance is insufficient. The reduction of the 
complex model does not produce rectangular rooms and obviously therefore 
the current version of UVAS does a very fine subdivision, which causes 
performance issues.

Complex dynamic geometry has been tested with UVAS unsuccessfully. In most 
cases the program crashes. Simple moving geometries have been shown the 
work properly. We are still waiting for the final version of UVAS, so we 
might have more on Monday.

5) The audio renderer works fine and is very stable. It has been tested 
with our speaker system and a large number of reflections. A late 
reverberation tail is added to the early reflections smoothly, and there 
is a novel system which defines the parameters for the late reverberation 
automatically.

6) We still do not have a model which could be used by all parts of the 
chain. The simplest models work, but they are visually dull. More complex 
models could not be utilized, although they went through reduction. We 
tried to create an ideal model by copying the furnitures from the newest 
model to the previous "test scene", but since we are not competent Max 
users it proved to be too difficult to do within reasonable time.

7) The whole chain has been tested. The stability is still an issue and 
we have found that the order in which the programs are started can be  
crucial for successful cooperation. Restarting the whole chain can take 
minutes, so it better to make sure that the data on the servers is not 
mutilated while playing with tools that can edit the data. Occasionally, 
some program drops from the chain which forces restarting.

Hopefully, we have a working demo on Monday!


Samuel Siltanen
saasilta at cc.hut.fi