Dialogue Voice

Workflow

1. Read core contract, CHARACTER_PROFILE.md, style rules, screenplay, 02_screenplay/line_manifest.jsonl when present, character bible, canon, and continuity.
2. Assign one VOICE-### identity per recurring voice. Use the canonical speech signature for cadence, habitual delivery behavior, and language style, while keeping acoustic timbre/accent/reference-audio identity in the voice bible.
3. Write 04_generation/voice_bible.md with vocal age range, register, texture, cadence, accent or dialect only when decided, speech habits, emotional range, and forbidden drift.
4. Write 04_generation/voice_cues.jsonl with the originating LINE-###, exact screenplay text, scene, speaker, language, delivery intent, timing target if known, measured duration when available, and target model.
5. Do not rewrite dialogue merely to make TTS easier without updating the screenplay.

Done

Every spoken LINE-### in scope maps to one voice identity and exact text source, and measured speech duration is available to the shooting script when generated.