3-D Audio Spatialization using Vector Based Amplitude Panning (VBAP) and Reverberation Cues



Site Map

Reverberation Cues

  • Two types of reverberation: localized and global
    • Global- Sent to every speaker.
    • Localized- Panned between a speaker triplet.
  • Gain of global reverberation is proportional to inverse of sound-source distance
  • Gain of localized reverberation is proportional to distance
  • As the sound source moves further from the listener, the reverberation becomes more localized.

VBAP

  • A generalization of the common cosine or tangent panning techniques.
  • Allows for simple computation of gain vectors for arbitrary geometries of multiple speakers,
  • Fundamental Concept: Express the location of a sound source in three-space as a linear combination of the vectors pointing from the listener to the speakers.
  • Details

Sound Examples

Extensions and Limitations

The biggest drawback to my implementation is that I require the user to define the speaker triples for a new group of speakers. This can be done automatically with a method desrcibed by Pullki in his paper. The problem is essentially one of finding a maximally connected graph on the surface of a sphere.

One shortcoming of Pullki's formulation is that he does not take into account the possibility that not all speakers are equidistant from the listener. In fact, his VBAP can very easily be generalized to speakers of varying distances from the listening position. Instead of scaling all speaker vectors to unit vectors, as Pullki does, one could normalize all of the speaker vectors by the shortest speaker vector.

One final limitation of both Pullki's and my implementation is that the system needs two different modes: one for a three-dimensional speaker geometry and one for a two-dimensional speaker geometry. I chose to implement the three-dimensional case, seeing as it is more general and it was possible to use it to for a two-dimenionsal geometry. In order to create the examples for the ballroom (a two-dimensional geometry), I defined a fifth speaker directly above the listening location. All of my sound location vectors were then defined to lie in the plane of the four speakers in the ballroom. After the usual calculations, I discarded the fifth channel (which contained only silence).


Last modified: Sat Dec 8 16:03:49 PST 2001