The description in the article sounds like they hooked the neurons to electrodes and fed the “voice” in as electrical impulses, and then used software to look at the neuron responses and determined they could tell which voice it was.

That doesn’t really seem very significant in itself. Of course if you give different inputs you’ll get different responses, and you can map those responses to those inputs when there’s no other complexity going on. It feels like a very very preliminary study being reported in a bit if a sensational way.