Generated audio

Operations (1)

• Create private audio

  speech.synthesize

  Create one owner-only MP3 from text, splitting long Azure batch input internally without trimming it. Put this operation on the flow path after the script-writing step; it cannot be an agent-selected tool. Use SKIP:no-profile, SKIP:no-eligible-sources, or SKIP:thin-material for an honest no-audio result.

  Inputs

  • text (string, required) - Original script text, or one supported SKIP code.
  • title (string) - Short episode title. Defaults to Generated audio.
  • delivery (string) - Closed speaking style. podcast is warm, polished narration and is the backward-compatible default; natural is conversational.
  • language (string) - Optional language name or BCP-47 locale. Blank keeps automatic source-language reading.
  • voiceGender (string) - Optional provider-neutral voice preference. Defaults to female.
  • pace (string) - Optional whole-recording pace: slower, normal, or faster. A subtle change that works for every supported language; normal means no rate change. Leave blank to keep the agent's current sound.
  • tone (string) - Optional whole-recording tone: natural, calm, energetic, or serious. natural means the voice's default reading. A tone other than natural works only when language is explicitly English; an unsupported choice fails before any audio is made and is never silently replaced.
  • sourcesJson (string) - JSON array of recorded-rights source records: sourceId, title, HTTPS url, fetchedAt, optional updatedAt, rightsBasis, HTTPS licenseUrl, and attribution.
  • sourceTextJson (string) - Optional bounded source-item JSON used only to reject long copied passages and thin material; source excerpts are never stored in the audio artifact.

  Returns

  • kind (string) - Stable generated-audio marker.
  • audio (object) - Owner-only MP3 media reference; absent on a skip.
  • aiGenerated (boolean) - Always true.
  • disclosure (string) - Runtime-owned spoken and visible AI disclosure.
  • title (string) - Episode title.
  • generatedAt (string) - UTC generation time.
  • scriptSha256 (string) - Digest of the disclosed spoken script.
  • sources (object) - Bounded source and rights records.
  • notice (string) - Owner notification text.

  Safety rules

  • The runtime adds the AI disclosure after model generation; never add or remove provider configuration.
  • Only recorded-rights HTTPS source records may be supplied.
  • Audio is private to the definition owner and never belongs on a public media route.
  • One run may make at most one speech call; there is no automatic retry.