Clone Voice

Clone a custom voice from an audio or video sample and get back its id for use in POST /v2/tts and generation text inputs.

The exact sequence for a local file: (1) request a presigned URL via POST /v2/assets/upload, (2) PUT the file bytes to the returned uploadUrl, (3) clone from the returned url (or register it via POST /v2/assets and clone from the assetId). Samples must be hosted in sync. labs storage — external URLs are rejected with errorCode voice_sample_upload_required. Video sources have their audio track extracted automatically (first 2 minutes). Video sources staged for extraction are limited to 1 GiB (1,073,741,824 bytes); larger sources return 413 voice_sample_source_too_large. This endpoint allows 10 requests per hour per API key (or per signed-in user).

For best results use 30 seconds to 2 minutes of clear, single-speaker speech without background music or noise. Samples that are too short are rejected with errorCode voice_sample_too_short. Clone slots are limited per plan (403 voice_clone_limit_reached) and voice names must be unique within your organization (409 voice_name_conflict). Sample source problems return 422 (voice_sample_not_accessible, voice_sample_type_unsupported, voice_sample_upload_required, voice_sample_extraction_failed); invalid or too-short samples return 400 (voice_sample_invalid, voice_sample_too_short). A 429 may mean the request-rate limit was exceeded or shared provider capacity is busy (voice_clone_busy); follow the response retry guidance. Provider outages return 503 (elevenlabs_service_unavailable).