Made in 10Docs
GuidesAPI
Characters

Cloned voices

Make a voice of your own from a few short recordings you have the right to use, with a consent statement, and give it to your characters.

Your organization can have up to five voices of its own: a few short recordings of a voice you have the right to use, a consent statement agreed to, and the voice is made. A cloned voice is your organization's alone and works like any of ours: set it as a character's voice_id.

Coming soonIn the dashboard

The dashboard's Voices page shows your cloned voices; cloning one there is coming soon. Today, clone through the API with a secret key.

Before you clone, show the person the consent statement, word for word, and have them agree to it. GET /v1/voices/consent answers the statement and its version:

GET /v1/voices/consent
{
  "version": "2026-11-05",
  "text": "I confirm that this is my own voice, or that the person whose voice this is has given me their permission to clone it and use it with AI characters on this service. I will not use it to deceive anyone, or to make it seem that a real person said something they did not say."
}

Send consent=true with the recordings, and the consent_version you showed. Without consent the answer is 400 consent_required; with an older version than ours it's 400 consent_required too: show the statement again. We keep the statement's exact words and version with the voice, who agreed (the account and its email, or the key), when, and a hash of their address. A deleted voice keeps that record.

The recordings

  • 1 to 5 files, each at least 4 KB and at most 10 MB (413 file_too_large). Through our public address the whole request must stay under 11 MB (below); the api alone would take 25 MB together.
  • Audio, recognised by its content whatever the file is called: MP3 (best, 192 kbps or more), WAV, M4A, OGG, WebM, FLAC or AAC. Anything else is 400 unsupported_file.
  • At least 5 seconds a file and 10 in all. One to two minutes of clear speech by one person, without music or noise, works best.
  • Never kept. The recordings are checked, used to make the voice, and dropped.

One request, at most 11 MB

Through our public address, talk.madein10.com, a request body is passed on up to 11 MB; a larger one is refused with a plain-text 413 before it reaches the api. So the real limit for one clone is 11 MB for all its recordings together, not 25 MB. MP3 is small and works best.

Clone a voice

POST /v1/voices as multipart/form-data:

Prop

Type

curl https://talk.madein10.com/v1/voices \
  -H "Authorization: Bearer $SECRET_KEY" \
  -F name="Hamid" \
  -F description="Calm, for support" \
  -F labels='{"accent":"Persian","gender":"male"}' \
  -F consent=true \
  -F consent_version=2026-11-05 \
  -F files=@sample-1.mp3 \
  -F files=@sample-2.mp3

The answer (201) is the voice. It's ready at once, or needs_verification when it has to be verified before it can be used: it's kept, but not offered to characters until then.

ClonedVoice
{
  "id": "vce_…",
  "voice_id": "c38kUX8pkfYO2kHyqfFy",
  "name": "Hamid",
  "description": "Calm, for support",
  "labels": { "accent": "Persian", "gender": "male" },
  "status": "ready",
  "error": null,
  "samples": 2,
  "character_ids": ["chr_…"],
  "consent": { "version": "2026-11-05", "by": "key_…", "email": "", "at_unix": 1793300000 },
  "created_unix": 1793300000,
  "updated_unix": 1793300000
}

voice_id is what a character's voice_id takes; it's null until the voice is made. status is ready, needs_verification or failed.

Give it to a character

PATCH /v1/characters/{id}
{ "voice_id": "c38kUX8pkfYO2kHyqfFy" }

Your ready voices show in the catalog's voices after ours, with kind: "yours" and their clone_id. Another organization never sees them.

Endpoints

With a secret key. In the dashboard, cloned voices are for owners and admins.

MethodPathBodyAnswer
GET/v1/voices/consent{ version, text }: the statement to show
GET/v1/voices{ data: [ClonedVoice], consent: { version, text }, limit: 5 }, oldest first
POST/v1/voicesmultipart, above201 ClonedVoice
GET/v1/voices/{id}ClonedVoice
PATCH/v1/voices/{id}{ name?, description?, labels? }ClonedVoice: our copy of its name and labels
DELETE/v1/voices/{id}{ id, deleted: true }

When it fails

StatusCodeWhy
400consent_requiredNo consent, or an older statement than ours
400unsupported_fileA file that isn't audio we take
409voice_limitYour organization already has 5 voices
413file_too_largeA file over 10 MB (a whole request over 11 MB gets a plain-text 413 before it reaches us)
422clone_failedThe recordings couldn't be made into a voice; the message says why in plain words
503voice_clone_unavailableCloning isn't available right now; try again later

Either way, nothing is kept and the slot stays free. A clone cut off on our side shows failed after 10 minutes; delete it and try again.

Deleting a voice

The voice is deleted where it was made first, then here. Characters speaking with it go back to their face's voice (voice_id: null). If it can't be deleted where it was made right now, the answer is 503 and the voice stays; try again.

On this page