Expressive mode lets the agent's voice perform delivery tags such as [laughs], [sighs] or [whispers]. The tags can come from your prompt, a static message, or replies the model writes. Voices that cannot perform tags have them removed before speech, so a tag is never read aloud.
Turn it on with Expressive mode under Speech Settings. It sets enable_expressive_mode: true on the agent.
Which voices perform tags#
| Voice | What happens to tags |
|---|---|
Premium voices (eleven_v4_turbo) |
Square-bracket tags are performed as audio tags. <emotion value="happy"/> is converted to [happy]. |
Cartesia with sonic-3 |
Emotion words become Cartesia emotion markup, and [laughs], [laugh], [laughing] and [laughter] become laughter. Other bracket tags are removed. |
| Every other voice and model | All tags are removed. |
Cartesia emotions recognised: happy, excited, sad, angry, calm, content, curious, surprised, sympathetic, apologetic, confident, frustrated, scared, enthusiastic, neutral.
With expressive mode off, tags are removed for every voice.
What changes in the prompt#
When expressive mode is on and the agent's voice can perform tags, Tried adds a section to the system prompt:
## Expressive delivery
You may add short audio tags to sound natural, for example [laughs], [sighs], [whispers] or [excited], placed right before the words they apply to. Use them sparingly.So the model starts using tags without you asking. When the voice cannot perform tags, this section is not added.
Tag palette#
When the toggle is on, the section shows the tags [laughs], [sighs], [whispers], [excited], [sad], [curious], [pause] and [clears throat]. Click one to copy it, then paste it into the prompt or a static message.
Examples#
In a static welcome message:
[excited] Hi Sam, great news, your order has shipped!In the prompt, to steer when tags are used:
## Style
- If the caller makes a joke, react with [laughs] before you answer.
- When you give bad news, start with [sighs] and speak gently.Fields#
| Field | Type | Default | Description |
|---|---|---|---|
enable_expressive_mode |
boolean | not set (off) | Perform tags on voices that support them; remove them elsewhere. |
voice_id, voice_model |
string | 11labs-Adrian, eleven_v4_turbo |
Must be a Premium voice (or Cartesia on sonic-3 through the API) for tags to be performed. |
curl -X PATCH https://api.tried.com.au/update-agent/agent_3f1c9a7b2d4e6f8091a2b3c4 \
-H "Authorization: Bearer $TRIED_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "enable_expressive_mode": true, "voice_model": "eleven_v4_turbo" }'Tips#
- Use tags sparingly. Constant laughter or sighing sounds unnatural.
- Standard voices read tags as plain text, so expressive mode needs the Premium tier. Premium costs 0.08 per minute for the voice line against 0.04 for Standard.