11 · Everyone Has a Voice
Small thing, enormous effect, and almost nobody knows it is here: the characters can talk out loud, and they do not all have to sound the same.
Craft can read dialogue aloud. Left alone, every single character inherits the Game Master's voice — so your entire cast becomes one person doing all the parts, which is fine for an audiobook and quietly terrible for a world.
11.1 · What happens instead
Characters are cast when they are created, in the same moment they get a name. The world picks a voice pool once, at installation, and each person is then matched inside it on gender and rough age.
It happens at creation because it has to. A character who has already spoken twelve times and then acquires a voice sounds like a different person arriving, and everybody notices.
11.2 · Here is the clever bit
The pool is chosen for sound, not for nationality.
The voices are organised by real language, because that is how the models were trained — but casting a Finnish voice for your dock crews is not saying they are Finns. It is saying their vowels sit forward and their consonants are soft. It is a decision about timbre, exactly the way an author picks a name for how it lands in the mouth rather than for where it came from.
Once you have internalised that, the whole feature stops being a localisation menu and starts being a casting session.
11.3 · Two pools is the move
If your world has two tongues — a settled one and a hill one, a court register and a dockside one, the people who were here and the people who arrived — cast them from two different pools.
The divide becomes audible before anybody explains it. Players hear that there are two groups several scenes before they are ever told there is a distinction, and they will start guessing which one somebody belongs to from the first sentence. That is worldbuilding you get for free, delivered through a channel nothing else in the system uses.
It is also completely deniable. Nobody has to be told it is happening for it to work.
11.4 · What it does not do
It does not change how anyone talks.
Accent, dialect, verbal tics, register, whether somebody trails off or talks over you, whether they get more formal under stress or fall apart into fragments — all of that is separate, it lives on the character, and it runs whether or not a voice is ever assigned. A character with no voice cast is still a specific person to read; they just sound like the narrator when read aloud.
And the two layers are genuinely independent. A soft-timbred voice attached to somebody whose speech is clipped and hostile is not a mismatch — it is a person, and it is usually more interesting than matching them up.
AT THE TABLE · Two people in a room should not be able to trade lines with the names stripped off and still be tellable apart by their grammar alone. If they can't be told apart, that is the thing to say out loud.
11.5 · Blank is a real answer
A world that wants one narrator throughout is not misconfigured. It is a different kind of book, and there is a long honourable tradition of exactly that.
The failure mode is not no voices. It is half the cast cast and half of it not, which reads as a bug even though nothing is broken — some people are themselves and some people are the narrator wearing their name.
Pick one. Either everybody gets cast, or nobody does. The middle is the only version that sounds wrong.