AI voices
Choose the voice your AI receptionist speaks with on calls, tune how it sounds, teach it how to read names, and clone your own voice.
Where to find it
Voice › AI voices
Open this screen
Steps
- In the sidebar, open Voice › AI voices.
- Read the banner at the top: Voice in use names the voice calls speak with now, and the line under it says whether live calls already use the voice chosen on this page.
- Narrow the list with Country / language, Gender and Search, then press Listen to a sample on a card.
- Press Use this voice on the card you want (workspace managers only). The card then shows Currently in use.
- Adjust Speed, Stability and Similarity under Voice settings and playground, press Listen with these settings, then Save settings.
- Optional: give a phone line its own voice under Voice per phone line, add words to the Pronunciation dictionary, or press Clone my voice.
Settings on this screen
| Setting | What it does | Default |
|---|---|---|
| Hear what callers hear | Plays the voice in use with its saved settings, as a call speaks it. | — |
| Country / language | Shows only voices for this language. Japanese is listed first. | Japanese (Japan) |
| Gender | Shows only female or male voices. | No preference |
| Search | Finds a voice by its name. | — |
| Include multilingual voices | Also lists voices that speak several languages. | On |
| Listen to a sample | Plays a short sample of that voice. | — |
| Use this voice | Makes this voice the workspace voice for calls. Managers only. | — |
| Speech model | For OpenAI voices: which OpenAI speech model speaks. | gpt-4o-mini-tts unless the deployment sets another |
| Voice per phone line | Gives one phone line its own voice. | Same as the workspace voice |
| Voice settings and playground | Speed, Stability and Similarity of the voice in use, a sentence to try, and Save settings. | Speed 1.00×, Stability 50%, Similarity 75% |
| Pronunciation dictionary | How the AI voice reads names, places and words. | Empty |
| Clone my voice | Records your own voice so your agent can speak in it. | — |
| Vapi voices | Vapi's voice library. Shown only once an operator has connected Vapi. | — |
| Hide / Show | Platform operator only: hides a voice from every workspace. | — |
Tips
- Leave the Search box empty and switch Include multilingual voices on to see every voice that can speak Japanese.
- /app/voice-playground opens this page at Voice settings and playground.
- When nobody has chosen a voice, the app speaks in the default voice, Eve: talking to the AI on screen, the Playground, the setup wizard and Read aloud. A voice you or your account chose is always kept.
- Each voice card shows only what is known about the voice: its name, language and style.
Good to know
- Only workspace managers can change the voice; everyone can listen to samples.
- A workspace can make 30 new samples an hour; a sample already made does not count again.
Related how-tos
- Choose the AI's voice Pick the voice your AI agent speaks with on calls for the whole workspace.
- Voice settings and playground Adjust how the voice in use sounds, try it with your own sentence and save the settings for calls.
- Pronunciation dictionary Teach the AI voice how to read names, places and words it gets wrong.
- Clone your own voice Record your own voice so your AI agent can speak on calls in it.
- Give a phone line its own voice Make one phone line speak with a different voice from the rest of the workspace.
- AI voices: fixing common problems Fix the messages the AI voices page can show.
- OpenAI voices and speech model Choose an OpenAI voice and the OpenAI speech model that speaks it.
- Use your cloned voice for calls, or delete it Make your agent speak in your cloned voice, choose the engine that holds the conversation, or delete the voice.
- Choose a Vapi voice Switched off until an operator turns it on Pick a voice from Vapi's voice library as the workspace call voice.
- One voice for the whole call Understand why a call never switches voice halfway through.
- Test a voice before callers hear it Hear a voice the way callers will, before and after you choose it.