Custom Voices
Custom voices let your agents speak in a voice you provide. Record or upload a short sample, pick the best ten seconds, and the resulting instant voice clone is immediately selectable wherever an agent speaks — phone calls, real-time voice, and audio responses.
What it's for
Built-in voices are fine for prototypes, but a customer-facing voice agent usually needs to sound like your brand — a specific person, tone, and accent. An instant clone captures that from a single short sample and behaves exactly like a built-in voice afterwards: pick it once on the agent and every spoken interaction uses it.
Navigating here
From the dashboard sidebar, open Custom Voices → Instant Clones. The list has the standard ownership tabs (Owned By Me / Shared With Me / Company Wide); clicking a voice opens its detail page, where you can rename it and manage tags.
Voice cloning is gated by your company's voice settings. If the Custom Voices section is missing from your sidebar (or the page says voice cloning is not enabled), your company doesn't have the feature turned on — ask your administrator.
Key concepts
Instant voice clones
An instant clone is built from a single ten-second segment of clean speech — no training job, no waiting. Quality tracks the sample: one speaker, no background noise or music, natural conversational delivery. Recording at least 30 seconds and then choosing the best ten gives the model more to pick from.
Using a cloned voice
Cloned voices appear in the agent (and multi-agent) editor's voice picker: choose the Medium voice model, and your clones are listed alongside the built-in voices under Voice Name. From there the voice applies everywhere that agent speaks — phone numbers, the test window's audio mode, and real-time voice sessions.
Sharing
Voices follow the standard sharing model: private by default, shareable with specific teammates or company-wide, with transferable ownership. Share a clone company-wide and every agent builder on the team can select it.
Fine-tuned voices
A second tier — fine-tuned voice clones trained on larger datasets for higher fidelity — exists in the product but is currently disabled in the builder UI. Instant clones are the supported path today.
How to create an instant voice clone
- On the Instant Clones page, click Create Instant Clone and name the voice.
- Provide the sample, either way:
- Record — capture audio from your microphone; aim for at least 30 seconds of clear, natural speech.
- Upload — drop in an existing MP3, WAV, OGG, or M4A file.
- On the trim step, drag the selector across the waveform to choose the best 10-second segment — the cleanest, most representative stretch of speech.
- Click Create Voice Clone. The clone is ready immediately.
- Open an agent, set the voice model to Medium, select your clone as the voice name, and use the test window's audio mode to hear it before going live.
Only clone voices you have the right to use. Cloning someone's voice without their permission is a misuse of the feature.
Sharing and access levels
Every voice clone starts private to its owner. Share it with specific teammates — each at their own access level — or company-wide at a single level for everyone. Voices use a simple slice of the standard access levels:
| Level | What it lets a teammate do |
|---|---|
| READ | Use the voice on their agents |
| WRITE / PUBLISH | Nothing beyond READ — voices have no draft or publish step |
| ADMIN | Share, delete, and transfer the voice |
The owner always has ADMIN implicitly; any ADMIN can transfer a voice clone's ownership.