Voice Cloning

Overview

A cloned voice is made from 10-15 seconds of one person speaking. Once it exists, any of your voice agents can speak with it: set the agent's voice to clone:<id>.

• Consent is required. Every clone records who confirmed the speaker's consent, when, and which version of the consent statement they accepted.
• The recording is not kept. CallMissed stores only a fingerprint (SHA-256) of the sample. The audio goes to the speech engine solely to create the voice.
• Paid plans only. Voice cloning is not available on the Free plan.
• Your clones are private to your account. Another account cannot list, use or delete them.

Voice cloning is being rolled out. Until it is enabled for your account, creating a clone returns 403 with "Voice cloning is not available on this account yet.", and an agent set to a cloned voice speaks with its model's default voice.

Authentication

Authorization: Bearer cm_your_api_key

Reading clones needs the bots:read scope; creating and deleting them needs bots:write.

Get the consent statement

Show this text to the person creating the clone, and send its version as consent_version. A request with any other version is refused with 422, so your app always shows the current statement.

curl https://api.callmissed.com/api/v1/voice/clones/consent \
  -H "Authorization: Bearer cm_your_api_key"

{ "version": "2026-10-08", "text": "I confirm that the voice in this recording belongs to ..." }

Create a clone

POST /api/v1/voice/clones takes multipart/form-data:

curl -X POST https://api.callmissed.com/api/v1/voice/clones \
  -H "Authorization: Bearer cm_your_api_key" \
  -F "sample=@asha.wav;type=audio/wav" \
  -F "speaker_name=Asha Rao" \
  -F "consent_attestation=true" \
  -F "consent_version=2026-10-08" \
  -F "provider=sarvam" \
  -F "language=hi-IN"

{
  "id": "4f0c6b1e-2d7a-4a53-9a51-6f5d2c1b9e10",
  "voice": "clone:4f0c6b1e-2d7a-4a53-9a51-6f5d2c1b9e10",
  "name": "Asha Rao",
  "provider": "sarvam",
  "language": "hi-IN",
  "gender": null,
  "status": "active",
  "consent_version": "2026-10-08",
  "consent_at": "2026-10-08T09:14:03Z",
  "created_at": "2026-10-08T09:14:03Z"
}

Sample length and languages

A sarvam clone can speak any of its languages, whatever language the sample was in.

Use a clone on an agent

curl -X PATCH https://api.callmissed.com/api/v1/bots/{bot_id}/config \
  -H "Authorization: Bearer cm_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{ "values": { "voice": "clone:4f0c6b1e-2d7a-4a53-9a51-6f5d2c1b9e10" } }'

• The clone speaks on its own engine: a cartesia clone uses sonic-3.6 and a sarvam clone uses bulbul:v3, whatever tts_model the agent has.
• It works on agents that use speech-to-text, an LLM and text-to-speech. It cannot be combined with a voice_tier or a speech-to-speech model; saving that combination returns 422.
• Saving a clone: id that is not one of your active clones returns 422.
• If a clone is deleted later, agents still set to it speak with their model's default voice. Calls are never dropped over it.
• Cloned voices are not available on the Managed Voice Agent WebSocket yet.

Speech in a cloned voice is billed per character on top of the agent's text-to-speech rate.

List and get clones

curl https://api.callmissed.com/api/v1/voice/clones \
  -H "Authorization: Bearer cm_your_api_key"

curl https://api.callmissed.com/api/v1/voice/clones/{clone_id} \
  -H "Authorization: Bearer cm_your_api_key"

The list returns your clones newest first (deleted clones are left out).

Delete a clone

curl -X DELETE https://api.callmissed.com/api/v1/voice/clones/{clone_id} \
  -H "Authorization: Bearer cm_your_api_key"

The voice is removed here and at the speech engine. The response status is deleted once the engine confirmed, or deleting while CallMissed keeps retrying in the background. Delete a clone whenever the speaker withdraws consent. When an account is erased, all of its clones are deleted at the speech engine too.

Errors