DIVA - SIGGRAPH 97 System Configuration
Due to the high performance requirements the
computation is spread across a network.
Click for bigger image (58 kb).
Musicians and instruments are animated in real-time.
Movements are read from an animation file, but since they
need to be synchronized with the music the control has to be done
in real-time. Also some additional movements can be triggered
during the show in random fashion or as a reaction
to user input.
Each of the musicians is a human model with over 60 degrees
of freedom (a true human model would have several hundred).
The instrument playing movements are calculated in advance.
One multi-CPU machine with extensive graphics hardware
is dedicated to the job of animation.
Rather than using conventional synthesis methods we have
adopted physical modelling synthesis for sound source
implementation where possible.
Currently implemented instruments are guitar, bass & flute which
are synthesized in a single workstation. This machine also creates the
external sound sources (traditional synthezisers)
and sends them all digitally to the auralization machine on separate
The direct sound and the first reflections coming to
the listener are computed using the image-source method.
The listener is considered to be where the camera is and can
naturally move freely around. The direct sound and image sources
are HRTF-filtered to produce an authentic 3-D sound scape in headphone
listening. Since the sound of
real spaces also includes diffuse ambient reverberation, a separate
algorithm is used to create late reverberation. Furthermore, the
varying absorption characteristics of different materials as well as
air absorption are taken into account in the real-time acoustics model.
In the Electric Garden's loudspeaker listening setup,
HRTF-filtering is not used
as it is physically impossible to deliver auralized sound
to many spots in the listening area (as in concert situation).
Instead simple panning of the image sources is used to give
some feeling of space.
The auralization and sound source computation is very heavy work and a
single twin-processor machine is dedicated to the work.
One processor does the auralizing while the other processor
calculates the image sources.
Conductor following is computationally
the least heavy of our system components in spite of the neural
processing. Hardware requirements are however not easily
very low latency networking is required. Thus the
following software runs in the same machine that does the
music synthesis to lower the audio latency which is the critical
part of our setup.
DIVA SIGGRAPH 97 page
This page is maintained by
The page has last been updated 2.9.1998.