OpenrouterSdk.Api.Speech (OpenRouter SDK v0.1.4)

POST /audio/speech — text-to-speech via the dedicated endpoint.

{:ok, mp3_binary} = OpenrouterSdk.Api.Speech.create(%{
  model: "<some-tts-slug>",
  input: "hello there",
  voice: "alloy",
  response_format: "mp3"
})

File.write!("hello.mp3", mp3_binary)

the response is the raw audio bytes — we do NOT decode json on this endpoint.

caveat

openrouter's /audio/speech accepts a fixed allowlist of model slugs that are NOT exposed via /models, so you can't discover them through the catalog. for catalog-driven tts, prefer OpenrouterSdk.speak/2 (in OpenrouterSdk.Audio) — it routes through /chat/completions with audio output and works against any model in OpenrouterSdk.Catalog.Models.tts_models/0.

Summary

Functions

create(payload, opts \\ [])

create_stream(payload, opts \\ [])

stream the audio response as a Stream of byte chunks. consumers who want pid / fun delivery pass :into exactly like the chat streaming api.

Functions

create(payload, opts \\ [])

@spec create(
  map(),
  keyword()
) :: {:ok, binary()} | {:error, OpenrouterSdk.Error.t()}

create_stream(payload, opts \\ [])

@spec create_stream(
  map(),
  keyword()
) ::
  {:ok, Enumerable.t() | reference() | term()}
  | {:error, OpenrouterSdk.Error.t()}

stream the audio response as a Stream of byte chunks. consumers who want pid / fun delivery pass :into exactly like the chat streaming api.