Voices define how manuscript text is spoken in generated audio. Start with a project narrator default before adding internal dialog, spoken dialog, book overrides, or manuscript-specific voices.
Render a short preview after voice changes so provider and pacing issues appear early.