Audio API

Audio uses the same API key and model IDs returned by GET /v1/models.

Key points

Speech is priced by Unicode characters.
Transcription and translation use multipart/form-data.
Duration is rounded up to seconds with a one-second minimum.
Failed requests are not charged.

Generate speech or transcribe audio#

Use the OpenAI-compatible audio routes with an API key. Speech returns an audio file; transcription and translation accept multipart audio and return JSON text.

POST /v1/audio/speech

curl https://api.xpncore.com/v1/audio/speech \
  -H "Authorization: Bearer $XPNCORE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"AUDIO_MODEL_ID","input":"Hello from XPNCORE","voice":"alloy","response_format":"wav"}' \
  -o speech.wav

Audio models are listed by GET /v1/models with audio.speech, audio.transcriptions or audio.translations.

Audio parameters#

Speech accepts JSON with model, input and voice. Upload routes use multipart/form-data with model and file; language and prompt are optional.

POST /v1/audio/transcriptions or /v1/audio/translations

curl https://api.xpncore.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $XPNCORE_API_KEY" \
  -F [email protected] \
  -F model=AUDIO_MODEL_ID \
  -F language=en

Audio pricing#

Speech is charged by Unicode character count using the model’s speech price per million characters. Transcription and translation are charged by measured audio duration, rounded up to whole seconds with a minimum of one second. Failed requests are not charged.

Audio errors#

Use a model that declares the requested audio capability. Invalid multipart fields, unsupported audio formats, insufficient funds and upstream failures return the standard error object; an upstream failure releases the hold.

Open the setup wizard Playground