Choose or clone a voice
Four stock voices, or record a clip and clone your own.
Voices live on Library · Voices. Clone one once and any agent can speak it.
The stock voices#
Four ship with every workspace:
| Voice | Character |
|---|---|
| Ryan | Dynamic, natural — the default |
| Aiden | Sunny and easygoing |
| Serena | Warm and mature, English with a Chinese accent |
| Vivian | Bright and quick, English with a Chinese accent |
Press Preview on any card to hear it, then Use this voice to give it to the focused agent. The card for the voice in use reads In use.
Clone your own#
- Press Add custom voice.
- Stay on Record here, or switch to Upload a file if you already have a clip.
- Press Start recording and read the passage on screen aloud, naturally. Allow the microphone when the browser asks.
- Watch the counter. Under ten seconds is too short to clone from; the recorder tells you when you have enough and warns you before it auto-stops at thirty seconds.
- Press Stop, then Replay to check it. Re-record if it is not clean.
- Name it and press Save voice.
What makes a good clip: one person, a quiet room, no music, ten to thirty seconds of natural speech. Background noise is cloned along with the voice.
To upload instead, give it a name and choose a WAV, MP3, M4A or WebM file under 20 MB.
Your recording stays private to your workspace, is stored encrypted, and only your agents can speak with it.
Remove a custom voice#
Press the delete control on a custom card. Bidirection tells you which agents speak with it today, and confirms that deleting removes it for the whole workspace. Those agents fall back to Ryan on their next call, and the uploaded clip is erased from storage.
Stock voices cannot be deleted.
Where else the voice matters#
The composer's Voice & delivery step also sets the Tone and the Greeting — the first line a visitor hears. Those live on the agent, not on the voice, so two agents can share a voice and still open differently.