🏷️ Audio Tags

Eleven v3 tags work best when they are sparse, lower-case, and placed at real performance shifts. Do not decorate every sentence; direct the voice only when delivery changes.

🎚️ Rules

• one boundary, one cue.
• two compatible cues are acceptable when they solve a real delivery need.
• put tags before the phrase they affect.
• leave factual labels plain unless clarity needs a tag.
• use punctuation and blank lines for rhythm before adding more tags.
• avoid raw SSML, markdown emphasis, or stage directions outside brackets.

🎭 Core Tags

• emotion: [happy], [sad], [angry], [excited], [nervous], [calm], [warm], [confident].
• delivery: [whispers], [shouts], [softly], [slowly], [clearly], [dramatic], [curious], [encouraging].
• reactions: [laughs], [chuckles], [sighs], [gasps], [clears throat].
• pacing: [pause], [long pause], [hesitates].
• sound: [applause], [ding], [drumroll], [thunder], [rain].

🧪 Combos

• host_open: [announcer, suspenseful].
• answer_list: [clearly, curious].
• reveal: [confident, warm].
• continuation: [pleased, encouraging].
• caution: avoid more than two cues in one bracket for canonical scripts.

🧠 Quiz Bridge

• Quiz Tags gives the exact WWTBAM-safe tag set.
• Prompt Stack shows the canonical German question/answer templates.

✅ Gate

• can the listener understand it first pass?
• did the tag change delivery without changing meaning?
• are facts and answer labels still neutral?
• would the line still work if the model under-obeys the tag?

---