UnderSky Docs

Transcribe audio

Send an audio file to UnderSky's OpenAI-compatible transcription endpoint, choose a model and response format, and handle minimum billing and recovery errors.

Use POST /v1/audio/transcriptions to turn an audio file into text. The endpoint accepts multipart form data and returns the upstream transcription response.

Send a transcription request

Include the audio file and a model. The default model is gpt-4o-mini-transcribe when you omit model.

curl -sS https://api.undersky.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer sk-your-api-key" \
  -F "[email protected]" \
  -F "model=gpt-4o-mini-transcribe" \
  -F "response_format=json"

The gateway forwards multipart fields such as language, prompt, and temperature to the configured transcription provider.

Choose a model

The endpoint supports these models:

ModelInput priceOutput price
whisper-1fixed request pricingfixed request pricing
gpt-4o-mini-transcribe$1 / 1M tokens$4 / 1M tokens
gpt-4o-transcribe$84 / 1M tokens$140 / 1M tokens
gpt-4o-transcribe-diarize$84 / 1M tokens$140 / 1M tokens

Every successful request has a $0.10 credit minimum. See pricing for the current public table.

Pick a response format

Use json when your application needs a stable object. Use the formats supported by the configured upstream when you need plain text or timestamps. Keep the response parser tolerant because provider response fields can differ by model.

Handle failures

  • 400 means the multipart body or model is invalid. Confirm that the file field has a filename and that the model is supported.
  • 401 means that the API key is missing or invalid.
  • 402 means that the account cannot reserve the minimum credit amount.
  • 429 means that the request rate is too high. Retry with bounded exponential backoff.
  • 5xx means that the upstream or gateway failed. Retry only when your job can be repeated safely.

Read the error code guide and rate-limit guide before adding workers.

Use text to speech to create audio, then send the file to this endpoint. For video files, first use the video extraction guide to obtain an audio track.

The media API overview compares transcription with image, speech, and video workflows.

On this page