#How do you get IA not to say the emotions in parentheses? I specifically told it in the prompt not

1 messages · Page 1 of 1 (latest)

buoyant glacierBOT
#
AI-generated Answer for samuel293527

Short answer: make sure you’re using Eleven v3 with square‑bracket audio tags, placed right before the words they should affect. If tags are being spoken, they’re not being recognized as tags.

Checklist to stop “[calm]” (and similar) from being read aloud:

  • Use Eleven v3 explicitly. In the API set model_id to eleven_v3, or in Agents select “V3 Conversational.” Other models don’t support audio tags. (help.elevenlabs.io)
  • Write tags in square brackets only, e.g., [calm], not parentheses. Place them immediately before or after the phrase; each tag influences roughly the next 4–5 words. (elevenlabs.io)
  • Use recognized tag styles. [calm] is a documented example and should work with v3. (elevenlabs.io)
  • If tags seem ignored/spoken, lower Stability from Robust to Natural or Creative so the model follows directional prompts more readily. (elevenlabs.io)

If after this the tag text is still read out, double‑check that the request actually reaches Eleven v3 (logs/config) and that no preprocessing step is stripping the brackets before TTS. (help.elevenlabs.io)

Sources:
How do audio tags work with Eleven v3? – ElevenLabs
Best practices | ElevenLabs Documentation
ElevenLabs Eleven v3 Audio Tags: Emotional Context in Speech
Best practices | ElevenLabs Documentation
How do audio tags work with Eleven v3? – ElevenLabs

frail needle
#

i don't use the api i use the agents conversational mode with v3 and i check the logs i see the [] but the agent speak them loudly

buoyant glacierBOT
#
AI-generated Answer for samuel293527

Short answer: in Agents, expressive/audio tags are only interpreted (not spoken) when the TTS model is Eleven v3 Conversational. Also, only certain tags are documented to work.

What to do

  • In your agent, set Agent Voice → Text-to-Speech model to “V3 Conversational.” Expressive Mode (which parses tags) is enabled by default only with this model. (elevenlabs.io)
  • Use documented tags. Examples: [laughs], [whispers], [sighs], [slow], [excited]. These affect roughly the next 4–5 words. Pause tags supported by v3 include [pause], [short pause], and [long pause]. (elevenlabs.io)
  • If you still see square-bracket tags in the transcript/logs, that’s expected; they’re part of the LLM output. If you want to hide them in the embedded widget transcript, enable widget.strip_audio_tags. This controls display only, not audio. (elevenlabs.io)

Notes

  • The docs show additional audio/emotion tags for v3 (e.g., [sad], [laughing], [whispering], [sigh]). Tag behavior is still under active development and results may vary. If you use tags outside the documented examples (e.g., [calm]), the model may not apply them. Try swapping to a documented tag with similar intent. (elevenlabs.io)
  • ElevenLabs’ blog also outlines recommended audio tags and usage patterns for v3. (elevenlabs.io)

If after switching to V3 Conversational and using the documented tags the agent still reads them aloud, share a minimal example in a new run; we can verify whether the tag is recognized or if another model/voice setting is being applied. (elevenlabs.io)

Sources:
Expressive mode | ElevenLabs Documentation
Expressive mode | ElevenLabs Documentation
March 2, 2026 | ElevenLabs Documentation
Text to Dialogue | ElevenLabs Documentation
ElevenLabs Audio Tags: More control over AI Voices
Expressive mode | ElevenLabs Documentation

frail needle
#

this is my setup already of course but doesnt change anything