< Back to IRCAM Forum

KNN - Speaker Array Localization / spatial aliasing problem

Spat
#1

Hello!

I am using spat for a permanent museum exhibit installation that includes a grid of 12 speakers - 3 trusses of four speakers each. spat5.pan~ is configured to use KNN with the following defaults:

  • /ramp/time 0
  • /neighbors/exponent 2
  • /neighbors/maxdistance 3.5 (speakers are about 10 ft apart)
  • /neighbors 3
  • /spread 0.

My spat abstraction is loaded into a poly~ instance due to the large number of sources that need to be spatialized - each one is using /source/1/xyz - I’ve validated that these sources move independently when addressed in the poly, and the speaker mapping is also correct in the space. I have also thoroughly tested my trigger system to rule out any bugs there.

I am spatializing some footstep sounds of varying sizes - high (think little critter), medium (think horse hoof), low (think thumping elephant foot). Each sound is an independent one shot trigger on a new poly target.

I am getting very good localization for the big footsteps, but when I trigger the medium and high footsteps I hear the sound almost splitting to speakers on opposite sides of the room. It doesn’t happen every time, but does happen consistently with that sources iwth higher frequency content and not at all with the big footstep samples. All sounds are mono 16 bit 48khz.

It’s as if the sound is bouncing back and forth across mirror images from the current position on the array. Weirdly, when I view the RMS meters in spat5 viewer, I don’t see the other speakers activating… but these are also very brief sounds, so may be too fast to see.

Any ideas for potential solutions or configuration that I may have missed would be so helpful!

Thanks in advance!

#2

Hello,

It’s hard to say.
I agree that visual monitoring with RMS meters in spat5.viewer might not be sufficient because their refresh rate may be too slow.
Maybe you could implement other ‘monitoring’ strategies (e.g., sfrecord the speaker feed signals) to check if there’s indeed sound “splitting to speakers on opposite sides” (shouldn’t be the case).

What you hear might also be some in-situ acoustical phenomena, like wall reflections or vibrations on the trusses ??

Finally, “/ramp/time 0” seems a little suspicious as this disables amplitude-crossfades when moving the sources – this might generate slight audio clicks/artefacts. But maybe you used others crossfade mechanisms in your poly~.

Best,
T.