Voice and speech
The voice the caller hears, its speed and tone, the greeting, and pronunciation of tricky words.
Pick a voice
The Voice page sets the voice the caller hears, plus its speed and tone. (What the agent hears is on Hearing.) Browse the list and press play on any voice before you commit -- a voice that reads well on paper can land wrong on a phone speaker.
A brand-new agent ships with no voice selected. The publish gate refuses until you pick one, so you never ship a voice nobody heard. The platform default is the ElevenLabs Australian voice Emma.
Emma is the platform-wide default and a reasonable neutral starting point, but pick the voice that fits the business rather than leaving it in place.
Speed
Playback rate from 0.5x to 2.0x in 0.1x steps. Around natural pace is usually best: too fast reads as rushed, too slow as robotic. Use Preview to hear the current speed and tone before you save.
Tone
An emotional colour for the voice. The available control depends on the engine behind the voice you picked.
Cartesia voices
The emotion control offers six presets.
| Value | Reads as |
|---|---|
| Neutral | Even, no colour |
| Content | Settled, at ease |
| Happy | Bright, upbeat |
| Surprised | Raised, reactive |
| Sad | Subdued |
| Angry | Sharp (rarely useful for a customer line) |
ElevenLabs voices
An expressiveness preset instead: Natural, Expressive, or Steady. A support line usually wants the even end of the range.
The greeting
The greeting is the first thing a caller hears. You write it on the Prompt page, next to the system prompt.
It is spoken word for word the instant the call connects, straight to the voice engine, skipping the model round-trip. That is what makes the open of a call instant and consistent.
Every new agent ships with a greeting built from the workspace name:
Thanks for calling Mitchell's Plumbing. How can I help you today?
If the workspace has no name yet, it falls back to a plain "Thanks for calling. How can I help you today?"
Turning it off
You can turn the fixed greeting off. The runtime then falls back to a greeting the model generates on the spot, which costs you the instant first speech.
A good greeting is short and names the business. That one line sets the tone for the whole call.
Pronunciation
A pronunciation entry teaches the voice how to say a word it would otherwise mangle: a brand name, a place, an acronym, or jargon. Add the word and how it should sound.
Pronunciation fixes what the agent says. Recognition hints on Hearing fix what it hears. A brand name or product range usually wants both. Both are agent-wide rather than per version.