Copy page

Docs

Voice Rights And Safety

Voice Rights And Safety

Soynade supports reference-conditioned text to speech. Reference audio remains a core API input, but you may submit it only when you have the speaker's authorization and the necessary rights to the recording.

The current policy is:

• Policy ID: voice_rights_safety
• Version: 2026-08-04
• Public policy: Voice Rights & Safety Policy

The workspace owner must accept the current policy before reference-audio production access is activated. This workspace-level prerequisite does not replace the per-request attestation described below.

Required Per-Request Fields

Whenever a text-to-speech request contains reference_audio, reference_audio_url, or reference_audio_base64, include both:

Code
{
  "voice_rights_attested": true,
  "voice_rights_basis": "authorized"
}

voice_rights_basis accepts:

Value	Use when
self	The person whose voice appears in the recording is authorizing its use.
authorized	You have that person's explicit written or recorded authorization.

The attestation is:

I confirm that I am the person whose voice appears in this recording, or that I have obtained that person's explicit written or recorded authorization to use their voice for synthetic speech. I also confirm that I own the recording or have all rights necessary to submit and use it for this purpose. I understand that unauthorized impersonation, deception, fraud, political-candidate or public-official impersonation, election deception, unauthorized robocalling, harassment, and circumvention of safeguards are prohibited.

Public availability of a recording does not, by itself, provide authorization. Keep evidence of the speaker's authorization and your rights to the recording. Soynade may request that evidence when investigating a complaint or suspected misuse.

JSON Example

Multipart Example

For a batch or JSONL batch upload, include the same two fields inside each text-to-speech item's body whenever that item contains reference audio.

Validation And Attribution

Soynade enforces this condition across synchronous, asynchronous, batch, and JSONL batch paths. A request is rejected with 422 before synthesis when reference audio is present and:

• voice_rights_attested is missing or is not true; or
• voice_rights_basis is missing or is not self or authorized.

If those fields are valid but the workspace owner has not accepted the current policy, Soynade returns 403 with detail.code: "VOICE_POLICY_ACCEPTANCE_REQUIRED". The owner can accept the displayed version from the API Keys page in Soynade Console; creating another key does not bypass this workspace-level gate.

Default packaged-voice requests do not contain reference audio and do not require these fields.

Production requests are attributable to a workspace, project, and API key. Soynade records the applicable policy version and rights basis for audit and safety purposes. Send only the public headers documented in this portal.