Docs

Audio

Text to speech (an audio file back) and speech to text (a multipart upload).

Text to speech

POSThttps://iranrouter.com/v1/audio/speech

bash

curl https://iranrouter.com/v1/audio/speech \
  -H "Authorization: Bearer ir-..." \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/tts-1", "input": "سلام، این یک تست است", "voice": "alloy"}' \
  --output out.mp3

The response is an audio file, not JSON, so the cost and request id arrive as headers: x-cost-rial and x-request-id. You are billed for the text you sent, because a TTS response carries no usage of its own.

Speech to text

POSThttps://iranrouter.com/v1/audio/transcriptions

POSThttps://iranrouter.com/v1/audio/translations

bash

curl https://iranrouter.com/v1/audio/transcriptions \
  -H "Authorization: Bearer ir-..." \
  -F file=@voice.mp3 \
  -F model=openai/whisper-1 \
  -F language=fa

The request is multipart/form-data, and your file reaches the provider untouched — only the model name is rewritten to the one the provider knows.
The model field is required — a 400 otherwise — and must come before the file in the form, which is what every SDK does.
The wallet hold is estimated from the file size and settled against the real duration once the provider answers.

Both surfaces honour Idempotency-Key — see the note on Embeddings.