Skip to main content
Use mode: "text_to_dialogue" with eleven_v3 or eleven_v4 to generate a conversation with multiple speakers. eleven_v4_turbo supports single-speaker Text to Speech only. Dialogue uses the same Create audio generation endpoint as speech. Set each speaker’s voiceId in dialogue, and omit prompt and controls.voiceId.
Choose voices from the voice library. A request supports at most 100 dialogue entries, ten distinct voices, and 2,000 characters in total, including whitespace and speech tags. Supported output formats are mp3_44100_128 and pcm_24000.

​
Estimate and retrieve audio

Send the same request body to Estimate audio cost before generation to see the current final USD estimate without generating audio. Creating an asynchronous generation returns HTTP 202. Use the returned task ID with Get audio generation until the task reaches a terminal status. Successful output is in outputs.audios; polling renews its URL during the seven-day retention period. Reuse the same Idempotency-Key and request after a timeout. A changed request under the same key returns 409. Continue polling tasks in reconciliation without resubmitting. See the Audio API guide for status, usage, errors, and delivery behavior, or use completion webhooks to receive results.