GoforIsrael 2016 tech conference set for debut in China, page-19

I don't think it is just the position or separation of the speakers.  
If you look at the presentation of 2012 it touches on this.  
http://www.audiopixels.com.au/index.cfm/news/presentations/merriman-presentation-feb-1-2012/
This is expanded significantly in the presentation of 2015:
"Fundamentally, Audio Pixels produces Sound-from-Ultrasound allowing the speaker to project a narrow beam of sound that can be heard only along the path of the beam (directional speaker)."
http://www.audiopixels.com.au/index.cfm/news/presentations/market-overview-november-2015/
In other words, and I'm speculating here, by adjusting the timing of the pulses going to each pixel this can be used to adjust the orientation of the wavefront of the sound wave that is being reconstructed from the discrete digital pulses from each pixel.  This gives a coherent sound "beam" that can be pointed wherever you want it.  A multiplicity of chips or chip arrays presumably allows a multiplicity of sound beams to be pointed at different targets.