Skip to content
Radient

Voice and sound through one API

Available

Generate a voiceover or the sound around it.

Use separate asynchronous workflows for character-billed voiceovers and second-billed sound effects. Poll or inspect the result, then retrieve an MP3 through a time-limited signed URL.

Used with Local Operator

Local Operator can expose provider-dependent voiceover and sound-effect tools when its configured provider supports them. The underlying Radient API also gives developers direct submit, status, result and cancel operations.

Create an API key (opens in a new tab)Compare the workflows

Availability

Available

Voiceover, voiceover-flash and sound-effects models

Voiceover billing

Per character

Text-to-speech request workflow

Effects billing

Requested duration rounded up to whole seconds

0.5 seconds bills 1 second; 4.01 seconds bills 5 seconds; requests 0.5–30 seconds

Result

Signed URL: 24 hours

Radient server retention: 72 hours; provider retention separate

Two request shapes, one asynchronous lifecycle

Voice and effects are not interchangeable presets. Each has its own input and billing unit, then shares the operational states needed to finish or stop a request.

API workflow / no generated sample

GET /v1/tools/audio/models

The live model listing returns voiceover, voiceover-flash and sound-effects. The workflows below show the request, status, result and cancellation contract for each audio type.

Workflow A

Voiceover

Submit text and voice parameters for asynchronous speech generation.

  1. 01 / SUBMITPOST /v1/tools/audio/voiceover with text and the selected voiceover model
  2. 02 / STATUSGET /v1/tools/audio/status with the returned request id
  3. 03 / RESULTGET /v1/tools/audio/result returns artifact metadata and, on success, a signed MP3 URL
  4. 04 / CANCELPOST /v1/tools/audio/cancel records best-effort cancellation; a running provider call may still finish
  5. BILLINGPer character / voiceover
  6. CONTRACTMP3 output; inspect status and the settled usage in the result

Workflow B

Sound effects

Submit a prompt and a requested duration for an asynchronous effect.

  1. 01 / SUBMITPOST /v1/tools/audio/sound-effects with a prompt and requested duration
  2. 02 / STATUSGET /v1/tools/audio/status with the returned request id
  3. 03 / RESULTGET /v1/tools/audio/result returns artifact metadata and, on success, a signed MP3 URL
  4. 04 / CANCELPOST /v1/tools/audio/cancel records best-effort cancellation; a running provider call may still finish
  5. BILLINGRequested duration rounded up to a whole second: 0.5 seconds bills 1 second; 4.01 seconds bills 5 seconds / sound effect
  6. CONTRACT0.5–30 seconds requested; measured MP3 duration can differ slightly

24-hour signed URL.
Shareable until its signature expires.

72-hour Radient server retention.
Download anything needed for longer. This retention limit does not cover provider copies.

Default provider retention.
On the Radient-served path, the provider retains prompt/text and generated audio under its own policy.

Mechanism

How it works

  1. 01

    Choose the capability and model

    Read the public audio model list, then choose voiceover or voiceover-flash for speech, or sound-effects for a prompted effect.

  2. 02

    Submit the typed request

    A voiceover supplies text and voice parameters and bills per character. A sound effect supplies a prompt and requested duration, rounded up to a whole second for billing.

  3. 03

    Track the asynchronous request

    Keep the returned request id. Use status and result rather than assuming the submission response contains finished audio; cancellation is available but best-effort.

  4. 04

    Retrieve and retain the MP3

    A completed result carries artifact metadata and a signed URL valid for 24 hours. Radient retains its artifact for 72 hours; download anything you need into your own storage. Provider copies are separate.

Concrete use case

Produce narration and a transition as separate billable jobs

A product demo needs a spoken introduction and a two-second transition. The voiceover job bills for the narration characters; the sound-effect job bills for the requested two seconds. Each gets its own request id and terminal result, so a failed effect does not force the narration to be regenerated.

The application can retain two MP3 artifacts, report their actual settled usage, and record that cancellation is best-effort when a provider call is already in progress.

Before you rely on it

Requirements, limits, security and cost

Requirements

  • A Radient API key and sufficient balance
  • Asynchronous request-id handling in the client
  • Provider-dependent tool availability in Local Operator

Limits

  • Sound-effect duration is 0.5–30 seconds
  • Results are MP3
  • A signed result URL lasts 24 hours and Radient retains its artifacts for 72 hours

Cancellation and logging

  • Cancellation is best-effort; an in-flight provider call may complete and be charged
  • On the Radient-served path, the provider retains prompt/text and generated audio by default under its own policy; zero-retention is not enabled
  • Do not put credentials or sensitive personal data into audio text or prompts

Cost

  • Voiceover bills per character
  • Sound effects bill the requested duration rounded up to a whole second, not measured output duration: 0.5 s bills 1 second and 4.01 s bills 5 seconds
  • Completed audio is charged normally; consult the console for exact settled cost
  • Radient's 72-hour retention limit applies only to its artifact store, not provider copies

Feature questions

The details that change the decision