elevenlabs.providers.sgit.ai / experiments / Multi-speaker dialogue
Multi-speaker dialogue
A question and an answer, in two voices, generated in one call so the pacing holds across the turn. Concatenating two separate generations does not give you this, and the difference is audible.
Your own full key, in your own browser, bounded only by your plan's monthly quota. That is pattern 0 with a ceiling: the narrow case where it is defensible — the key's owner, testing their own key, on their own device — and not a pattern to publish. No key ships in this page: the field below is empty until you fill it.
ElevenLabs offers no per-key spend limit for text to speech, so nothing here caps what a leaked key could cost you except the account's own quota. Use a key scoped in the dashboard to the endpoints this lab needs, and forget it when you are done. The version of this page with no key box at all is pattern three, and it does not exist yet.
1 · The exchange
One request, several voices, consistent pacing across the turns — which is the thing you cannot get by generating each line separately and concatenating. A two-voice cut (a question and an answer) is one endpoint away from a single-narrator reel.
2 · The result
If this endpoint is not on your plan the call comes back 402 or 404 — that is the vendor answering a question about your account, not a bug in this page. Read the log.
3 · Log
What this tests
POST /v1/text-to-dialogue takes inputs: [{ text, voice_id }, …] and renders the whole exchange in one request with consistent pacing vendor docs 5 Sep 2026. The /with-timestamps variant returns the same character-level alignment as ordinary speech, so everything the captions studio does works on a dialogue too.
For a single-narrator explainer this is one endpoint away rather than a rewrite, and it changes what a script can be: a rhetorical question that is actually asked by somebody else lands differently from one the narrator asks themselves.
What to listen for
- The seam. Generate the same two lines separately, concatenate them by hand, and compare. If the single-call version does not sound more like a conversation, the endpoint is not earning its keep for your material.
- Turn-taking pace. Real exchanges overlap slightly and vary their gaps; a model that leaves an identical pause after every turn sounds like a phone tree.
- Whether the voices stay distinct when the text is technical and both speakers use the same vocabulary.
Cost and availability
Billed as characters, at the model's rate — the same arithmetic as ordinary speech vendor docs 5 Sep 2026. Availability is per-account: if the endpoint is not on your plan the call comes back 402 or 404 vendor docs 5 Sep 2026, which is the vendor answering a question about your subscription rather than a fault in this page. The log shows the status and the detail body, which is the useful part.
Nobody here has run it written, not run.