Create Speech
Convert text to speech.
Provider is resolved automatically from the model name via the model_offerings
table — no hardcoded provider branching. Streaming is used when the provider
adapter supports it and stream=true (the default).
The voice field is required for ElevenLabs models. Sarvam models use
speaker instead.
Authorizations
Enter your MeshAPI key (rsk_...) or, for the admin-keys endpoints, a dashboard session token — sent as Authorization: Bearer <token>.
Headers
Dated version of the API contract to pin this request to. Omit it and the request is served under 2026-08 — the oldest supported version, so an existing integration is never moved by a release. A malformed or unsupported value is rejected with 400 invalid_api_version rather than falling back silently. The version actually served is echoed as X-Mesh-Version on every response, including errors.
2026-08, 2026-09, 2026-11 Body
Output sample rate in Hz. Honoured on Sarvam-, Runware- and SiliconFlow-served models, and accepted but ignored on the others. Each model publishes its own accepted rates, and some accept fewer depending on response_format; Mesh forwards your value as sent and never substitutes a supported rate for one a model does not accept. It applies to the model you named: if your request is served by a fallback instead — which X-Mesh-Routing-Fallback: true on the response tells you — the rate is not applied, because the fallback may accept a different set.
Speaking rate, where 1.0 is the voice's natural pace. Honoured on DeepInfra-, SiliconFlow- and Venice-served models and on Novita's MiniMax speech models, and accepted but ignored on the others — Novita's Fish Audio model reads no rate control at all, and on ElevenLabs the rate control is voice_settings.speed instead, with this endpoint's default model being an ElevenLabs one. A model's own accepted range may be narrower than the one allowed here; Mesh forwards your value as sent and never clamps it.
0.25 <= x <= 40 < x <= 10happy, sad, angry, fearful, disgusted, surprised, calm, fluent, whisper Pin this request to one upstream provider, e.g. elevenlabs. The request is served by that provider or refused — never routed elsewhere. See GET /v1/providers/pinnable for the slugs this deployment accepts.
Your own labels for this request, echoed back on the usage row and available as a filter and a group-by dimension in the usage API. String keys to string values. Never put personal or sensitive data here — tags are stored with the usage record and are not redacted.
A stable, anonymised identifier for the end user making this request, recorded on the usage row and available as a filter and a group-by dimension in the usage API. Supersedes the deprecated user field. Use an opaque id, not an email address or a name.
256Response
Successful Response