It's basically a NeoPixel ring and an FPGA that handles 8 I2S microphones, plus some GPIOs and an optional ESP32 that can make the hat work stand-alone - but you still need a Raspberry PI to program it. Via SSH over another computer.
It's a BlinkenLights microphone.
It only get's interesting when you combine it with voice recognition software, granted the examples also work with a USB microphone connected to the Raspberry PI.
First I tried to get the wakeword example from their "matrix core" library to work, but seemed to have failed at setting up the Sphinx Knowledge Base. It triggered very randomly and never did I see a word that I had said, even if I said the word that supposedly triggered before that.
The second offline voice recognition software that I've tested is "porcupine" by picovoice and their speech-to-intent engine called "rhino" - pretty impressive stuff, but their context demos are for a coffee-machine, a music player _or_ a light switch. Nothing interesting for a robot.
Tomorrow I will "play" with the snips platform, as someone on twitter suggested.