Your own voices
Besides the voices in the voice list, an agent can speak with a voice of your own. You add it in two ways, under Settings, Your voices. It appears only in your workspace, marked as yours, and works like any other voice in an agent's voice settings: pick it, preview it and save the agent.
- Clone a voice. Upload or record a short sample of someone speaking, confirm they agreed, and zlicx makes a copy of the voice with ElevenLabs, Sarvam, Cartesia or Smallest.
- Import a voice. Paste the ID of a voice you already have in your own ElevenLabs, Cartesia or Smallest account.
Clone a voice
- Open Settings, Your voices, choose Add your own voice, then Clone a voice.
- Pick the provider and the languages the voice should speak.
- Upload or record the sample. One person, speaking clearly, with no music or background noise.
- Enter the speaker's name and confirm the consent statement. A signed-in member of your workspace must confirm it; API keys cannot.
Each provider takes a different sample:
| Provider | Length | Largest file | Formats |
|---|---|---|---|
| ElevenLabs | 10 seconds to 3 minutes (1 to 2 minutes is best) | 10 MB | MP3, WAV, WEBM, OGG, FLAC, M4A |
| Sarvam | 5 to 30 seconds (10 to 15 is best) | 10 MB | WAV, MP3, FLAC, OGG, AAC |
| Cartesia | 5 to 60 seconds (10 is enough) | 16 MB | WAV, MP3, FLAC, OGG, WEBM |
| Smallest | 5 to 15 seconds | 5 MB | MP3, WAV, MP4, WEBM |
zlicx checks the size and the real format of the file, and the length of WAV files, before anything is sent. The provider checks the rest and its reason is shown if it refuses the sample.
Consent
Only clone a voice whose owner has agreed. With each clone zlicx records the speaker's name, who confirmed their consent and when, the version of the statement they confirmed, and the address the request came from. The sample is kept privately as evidence for as long as the voice exists, and deleted with it. If the speaker withdraws their consent, delete the voice.
Import a voice from your account
- Save your provider's API key under Settings, Your AI provider keys.
- In Your voices, choose Add your own voice, then Use your own voice ID, and paste the voice ID from your provider account. zlicx checks with your key that the voice exists.
Smallest voices must be clones on its Lightning v3.1 Pro pool. Voices already in the zlicx voice list cannot be imported; pick them there.
Which key a call uses, and billing
| Voice | Provider key | Speech billed by |
|---|---|---|
| Cloned | zlicx's, even if you saved your own key for the provider | zlicx, per character at the model's usual rate. Cloning is free. |
| Imported | Always yours | Your provider, directly. zlicx charges nothing for the speech. |
If you remove the provider key an imported voice needs, calls and previews with that voice are refused until you add the key again or choose another voice.
Limits
A workspace can have 3 voices of its own (cloned and imported together). Contact zlicx support to raise the limit. The limit and how many you use are shown with your other workspace limits on the billing page.
Delete a voice
A voice that an agent uses cannot be deleted; choose another voice for those agents first. Deleting a cloned voice removes it from the provider and deletes its sample. Deleting an imported voice removes it from zlicx only; it stays in your provider account.
When a workspace is erased, its cloned voices are deleted at the provider together with their samples. A workspace export includes each voice's details and consent record, not the audio.
API
The same actions are available under /api/v1/voices/custom with an API key that can configure agents. Cloning needs a signed-in member, because it records their consent.
# Import a voice from your Cartesia account
curl -X POST https://in-api.zlicx.ai/api/v1/voices/custom/import \
-H "x-api-key: $ZLICX_API_KEY" \
-H "content-type: application/json" \
-d '{"provider":"cartesia","providerVoiceId":"<voice id>","name":"Support voice","languages":["en","hi"]}'Your voices also appear in GET /api/v1/agents/voice/catalog with custom: true, and an agent uses one by its voiceId like any other voice.

