Audio

Planck exposes explicit OpenAI-compatible audio routes.

Capability	Route
Transcription	`POST /v1/audio/transcriptions`
Translation	`POST /v1/audio/translations`
Text to speech	`POST /v1/audio/speech`

Transcription

curl https://api.inquantum.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $PLANCK_API_KEY" \
  -F model="gpt-4o-transcribe" \
  -F file="@meeting.mp3"

Translation

curl https://api.inquantum.ai/v1/audio/translations \
  -H "Authorization: Bearer $PLANCK_API_KEY" \
  -F model="whisper-1" \
  -F file="@spanish-audio.mp3"

Speech

curl https://api.inquantum.ai/v1/audio/speech \
  -H "Authorization: Bearer $PLANCK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini-tts",
    "voice": "alloy",
    "input": "Hello from Planck.",
    "instructions": "Speak warmly and clearly.",
    "response_format": "mp3"
  }' \
  --output speech.mp3

Use the bare OpenAI model ID on this route. The response contains audio bytes; save them to a file or play them in your application. GPT-4o Mini TTS billing uses the terminal speech.audio.done usage: input tokens are text input and output tokens are generated audio. Both contribute to the final cost. Missing usage remains incomplete rather than a certified zero charge. When private request-asset storage is configured, generated speech is saved with the request using the existing image-asset storage and signed-URL access path. The request view provides playback and download. Send Planck-Omit-Response: true to omit the response body and audio asset from logs; usage and billing are still recorded. Planck-Omit-Request: true omits the input text and instructions. Historical logs that recorded only audio body size cannot recover the audio.

BYOK

Audio follows the same BYOK/PTB selection as chat: stored BYOK is used by default when available, otherwise PTB is used when available. Send Planck-Billing-Mode: ptb or Planck-Billing-Mode: byok to force a mode for a specific request. X-Provider-Key and Planck-Provider-Key are still supported as explicit per-request BYOK overrides. Audio requests are usage-bearing and should flow through gateway cost extraction, billing, and logging when the provider returns enough usage information.

This endpoint uses the shared AI Gateway error envelope. See the Error Handling & Fallback guide for response examples, retry rules, fallback behavior, and streaming errors.

Gateway API

Observability API

Prompt API

Transcription

Translation

Speech

BYOK

​Transcription

​Translation

​Speech

​BYOK

Transcription

Translation

Speech

BYOK