>That doesn't sound like the same bar as "most authentic full-duplex speech model ever built"

It's up there. I think most humans would be convinced they're talking to a human. I assumed you've used other speech models. They are okay, but not as natural and imperfect as Dessa is.