Stream synthesized speech
OpenAI-compatible text-to-speech. Use Authorization: Bearer <API_KEY>
and a JSON body. The response is a raw audio byte stream forwarded
progressively (no prediction polling); the negotiated content type is
returned verbatim (audio/mpeg for mp3, audio/pcm;rate=…;channels=…
for pcm). An omitted response_format defaults to mp3.
The accepted model/voice/format set is a curated per-environment
catalog that a model joins only after passing contract qualification;
deepgram/aura-2 is the launch model. An unsupported model, voice, or
format, or a speed/instructions value for a model that does not
support it, returns a 400 without dispatching a synthesis.
This endpoint is independently feature-gated during preview rollout. A disabled deployment returns an OpenAI-shaped 404 without dispatching.
Authorizations
API key passed as an Authorization Bearer token: Authorization: Bearer YOUR_API_KEY
Body
Must be in the environment's speech catalog. An unlisted model returns 400 without dispatching a synthesis.
"deepgram/aura-2"
Text to synthesize. Bounded by the model's per-character limit (counted in Unicode runes).
Model-specific, namespaced voice id. Must be allowlisted for the model.
"aura-2-thalia-en"
Audio format. Omitted defaults to mp3. Allowed values are per-model (e.g. mp3, pcm).
"mp3"
Playback speed. Accepted only for models that declare support; otherwise 400.
0.25 <= x <= 4Steering instructions. Accepted only for models that declare support; otherwise 400.
Only raw audio streaming is supported. sse is rejected with 400.
audio Response
Raw audio byte stream. Bytes are forwarded progressively as the
provider synthesizes them; the content type is the negotiated
format (audio/mpeg or audio/pcm;rate=…;channels=…).
The response is of type file.