That is a very good question because currently I feel there isn’t a product that is in the right price range.
If there was product available I would be trying to adopt a tiered approach where 2mic/speaker satelites of low cost with vad/aec.
These are for multichannel wide area room microphone arrays, so each device is essentually mono procesing a channel, so cheap, really cheap as opposed to the respeaker.
Currently small singular array microphones are packing much DSP load with need for beamforming whilst multiple units with just vad/aec maybe NPU wakeword acceleration is all that is needed.
In a tiered system the satelites would stream channels to zone authoritive servers with a more powerful NPU and possible DSP for beamforming.
In a wide array microphone system channel seperation and timing is far more pronounced so maybe DSP is less of an issue.
I am not really sure where Arm are pitching this. If I use Rockchip somewhere in the middle of a RK3308 and RK3399pro, which may be OK for authoritive zone servers but likely expensive for satelites simple arrays.
Allwinner have 2 interesting lowcost chips the R328 & R329 the Chinese even have $10 product with the Tmall Genie, so something much cheaper is needed.
The current respeaker provide far field high quality audio in the format of a conference table microphone and its likely that level isn’t needed for near field distributed recognition quality systems.
Optimised 2 Mic systems that are Pi4-2gb price or less.