Voice cloning API

Send POST /voices with voices:write scope and kind: "clone".

curl --fail-with-body https://app.verbatik.com/api/v1/voices \
  -H "Authorization: Bearer $VERBATIK_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Idempotency-Key: narrator-voice-001' \
  -d '{
    "kind":"clone",
    "name":"Campaign narrator",
    "audio_url":"https://example.com/narrator.wav",
    "preview_text":"Welcome to our latest collection.",
    "noise_reduction":true,
    "volume_normalization":true
  }'

Replace the example URL with a reachable public HTTPS recording, or upload a supported file through POST /files. Use one speaker with permission to use their voice.

Fields and result

name is required and limited to 80 characters. audio_url is required. preview_text is optional and limited to 500 characters. Noise reduction and volume normalization default to false.

Unlike queued media generation, successful voice creation returns 201 with the created voice, its id, preview information, and credits. Allow enough client timeout for the operation to finish. Retry uncertain requests using the same idempotency key.

The charge includes voice creation and preview synthesis. Later speech requests using that voice are charged separately. There is no fixed $3-per-clone promise in the current credit-based API.

Use the returned voice resource ID with the Speech Cloning model in POST /speech. The voice.created webhook can notify your integration when a voice is created.