Speaker label: naming who talks inside a prompt
A speaker label is a name written beside a line so the model knows which character says it. One entry documents the convention. Inside a call it does real work; across calls it does nothing at all. As of 2026-09-12.
| What it does | Inside one generation | Across generations |
|---|---|---|
| Routes a line | Sends each line to the right face | Nothing published |
| Orders an exchange | Turns can be alternated deliberately | Nothing published |
| Establishes a voice | Consistent within the call | No mechanism connects a label to a voice |
Inclusion rule. The functions a documented speaker label performs, separated by whether the documentation extends them beyond a single generation. Order. Fixed order: routing, ordering, then identity.
1A label is an instruction, not an identity
Labelling speakers consistently and alternating their turns is the only published guidance in this register on the commonest shot in drama. It is genuinely useful and it operates inside one request.
A two-hander can therefore play correctly in one shot and come back with different voices in the reverse angle. For a serial that is the gap that matters, and no sentence on either page closes it.
2Consistent labelling is still worth doing
Even without persistence, naming the same character the same way keeps a shot list readable and keeps model behaviour predictable inside a call. It costs nothing and it is the shape a voice mechanism would want if one ever arrived.
What it cannot do is substitute for a cast. A serial needs voices that outlast a request, and a label is not one, however carefully it is written.
3A label and an identifier are different tools
An identifier points at a stored voice and returns the same one every time. A label points at a character inside a piece of prose and returns whatever the model decides that character sounds like on this occasion.
Both are passed on a request and only one of them is a casting decision. Treating them as equivalent is the commonest way a production over-estimates what a prompting guide has promised.
4Sources read for this entry
This page defines the term and logs what each vendor documents about it, each figure read on 2026-09-12. Related: Voice catalogue, Language code.