What would make it "natural-sounding", though? What if that happens to include gender stereotypes?

I don't think it's possible to assert that it's purely a technical challenge before we understand the goal clearly, beyond "I'll know it when I hear it".