Content Policy

Overview

Text submitted for synthesis is screened against Speechify’s content policy. A request refused under that policy returns 400 content_policy_violation, or, on a streaming request that has already started playing, ends early. A refused request is never billed to your account.

Screening applies to every surface that renders text into speech, including multi-speaker synthesis:

What a refusal looks like

{
  "error": {
    "code": "content_policy_violation",
    "message": "This request was refused by the content policy (<category>). Edit the input and try again."
  },
  "request_id": "..."
}

Treat it as a persistent error: the same request will be refused again. Editing the input is the only thing that changes the outcome, so do not retry it unchanged and do not treat it as a transient failure. See Error Handling for the retry rules on other statuses.

Streaming: a refusal can end a stream early

Screening runs alongside synthesis, so on the streaming endpoints audio can be ready before the screening result is. A refusal arrives as a 400 before any audio when screening finishes first. That includes a text your workspace already had screened, because a result is kept and reused for 24 hours after it is first stored. Otherwise the stream starts, and about the first half-second of audio can play before the result is in; on Opus and AAC output the stream always waits for the result.

When screening is still running as the audio starts, a refusal that arrives while the stream is playing ends it:

Build against these rules:

• Only a stream that completes normally (the transfer finishes, or speech.done arrives) was served in full. Treat an aborted /v1/audio/stream transfer as a failure and discard the partial audio.
• A speech.error with content_policy_violation is the same persistent refusal as the 400: edit the input rather than retrying.
• A stream ended by a refusal is not billed.

What is prohibited

You may not use Speechify to synthesize content that, among other things:

• sexualizes minors;
• threatens violence against a person or a group;
• provides instructions for self-harm;
• is likely to be used as part of a financial scam or other fraud;
• impersonates a public official or a candidate for office.

The full rules are in the Terms of Service, which govern.

These are prohibitions on your use, and they are broader than what any automated screen can recognise. Screening is one enforcement mechanism and is not a guarantee that prohibited content will be detected in every case; you remain responsible for the content you submit, Speechify reserves the right to refuse any request under the policy, and repeated abuse is handled at the account level up to and including termination.

Working with the policy

Most published work is unaffected. Fiction, journalism, true crime, history and court reporting routinely describe or quote violence and threats, and depicting or reporting on this material is treated differently from producing it. If legitimate content is being refused, contact us with the request_id from the response rather than rewording around the refusal - the request_id is what lets us look at the specific decision.