Capability

Voice Cloning

Create a reusable private voice clone from one consented reference recording. The request queues a clone task; poll the completed task to obtain its saved voice ID for later generation flows.

Quickstart
Use the saved voice
Request fields
Consent records
Pricing

Dedicated route only

Voice cloning uses POST /v1/voice-clones. It is intentionally not exposed through POST /v1/generations or a compat route because the API creates an auditable consent record before queueing the worker task.

Quickstart

POST/v1/voice-clonescreate a reusable voice clone

curl -X POST https://apiv2.soundverse.ai/v1/voice-clones \
  -H "Authorization: Bearer sksoundverse_..." \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: voice-clone-req-001" \
  -d '{
    "reference_file_id": "018f0000-0000-7000-8000-000000000001",
    "name": "Lead vocalist reference",
    "description": "Approved reference recording for internal demos",
    "version": "v5",
    "license": "royalty_free",
    "consent": true,
    "rights_attestation": "We have consent from the speaker and rights to use this recording for voice cloning."
  }'

The reference_file_id must be an active uploaded audio file owned by the authenticated enterprise account. For best results, use a clean 15–30 second mp3 or m4a recording of one speaker, with little or no music, reverb, or background noise. The API accepts reference files up to 15 MiB.

Read the saved voice ID after completion

The create response is {"task_id": "...", "status": "queued", "consent_id": "..."}. Poll GET /v1/generations/{task_id}; once completed, read the saved Soundverse vocal_id from output.metadata_json. Parse that JSON and copy itssv_voice_... value into the next song or singing request—never use the provider voice ID.

The clone task creates a reusable voice record, not an audio download. Audio generated with that saved voice is later available in the same account’s Soundverse Library. Use GET /v1/files/{file_id}/download for API-only download. Enterprise API outputs are hidden from public/profile surfaces by default.

Use the saved voice in generation

Voice cloning creates a reusable sv_voice_... ID; it does not create a separate inference endpoint. After the clone task completes, pass that saved vocal_id to an existing song or singing generation request.

POST/v{5,6,7}/generate/singingsing using the saved voice

POST/v{5,6,7}/generate/songgenerate a song using the saved voice

curl -X POST https://apiv2.soundverse.ai/v7/generate/song \
  -H "Authorization: Bearer sksoundverse_..." \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: song-with-cloned-voice-001" \
  -d '{
    "prompt": "warm cinematic pop with piano and strings",
    "lyrics": "Walking through the city lights, I hear the morning call.",
    "vocal_id": "sv_voice_..."
  }'

The example sends the cloned voice to POST /v7/generate/song. To make a vocal-only result, use the same vocal_id with POST /v7/generate/singing and replaceprompt with styleOfMusic.

The saved voice must belong to the API-key owner and have active consent. Send the Soundversesv_voice_... value only—never a provider voice ID. Saved voices work with prompt-based and reference-song generation, but not with melody_url. See Singing Generation and Song Generation for complete request and result details.

Request fields

Field	Type	Required	Default	Description
reference_file_id	uuid	Yes	—	Owner-scoped uploaded audio file to clone from. Must be mp3/m4a and 15 MiB or smaller; a clean 15–30 second solo-speaker recording is recommended.
name	string	No	null	Optional display name for the saved voice. Maximum 255 characters.
description	string	No	null	Optional description stored with the saved voice. Maximum 1024 characters.
version	"v5" \| "v6" \| "v7"	No	"v5"	Voice clone model version. v6 and v7 may require a higher plan.
license	license tier	No	"royalty_free"	License tier to price and attach to the queued clone task.
consent	true	Yes	—	Must be true. Confirms the referenced speaker consented to this clone.
rights_attestation	string	Yes	—	Recorded statement that the caller has the necessary voice and recording rights. Maximum 2048 characters.

Pricing

$12.00 USD per voice clone. The price is flat across v5, v6, v7, and all license tiers.

Version	Price
v5 / v6 / v7	$12.00

USD flat price per clone, regardless of license tier.

← RemixAudio Inpainting →

Voice Cloning

Quickstart

Use the saved voice in generation

Request fields

Consent records

Pricing