But there are already voice models that do a reasonable job at a fraction of the throughput available?

The real question is how expensive it is to coordinate between these different modalities, and I really don't see why it'd be all that much.

I half expect Boston Dynamics to show something like this off in Q4 or whatever.