Docs
Voice and speed
Voices
Find a voice that fits your business. Preview it, choose it or clone your own.
Voices come from Cartesia, ElevenLabs, Deepgram and OpenAI. The picker ranks them by how human listeners rated them and lets you preview each one with your own greeting.
| Provider | Model | Reply latency | Notes |
|---|---|---|---|
| Cartesia | Sonic 3 | about 110 ms | Fast and natural. The default for most assistants. |
| ElevenLabs | Flash v2.5 | about 150 ms | A large library of voices and accents. |
| Deepgram | Aura-2 | about 280 ms | Clear voices at a low cost. |
| OpenAI | GPT-4o mini TTS | about 480 ms | Expressive, but slower to start speaking. |
#Which voices you can use
Core includes 25 selected voices. Pro includes 400+ voices. See current plans for details.
#Speed
Set the speaking speed from 0.7× to 1.2×. Slightly slower often suits older callers or complex details like addresses.
#Clone your own voice
On Pro, owners and admins can clone a voice from a recording between 10 seconds and 3 minutes long (up to 12 MB). You must confirm that the person recorded agreed to it. Cloned voices appear first in the picker and can be deleted at any time.
- Record in a quiet room, close to the microphone.
- Speak naturally, the way you would answer the phone.
- A minute of clean speech is enough.
Note
Ultra mode uses its own set of voices, because one model hears and speaks.