Skip to main content
Transcribe audio files into timestamped text segments with optional speaker diarization. Ideal for podcasts, interviews, and recorded calls that need to be searchable or summarized.

​
Request body

  • sourceUrl — HTTPS or signed URL to the audio file (MP3, WAV, FLAC, etc.). Required if file is not provided.
  • file — Optional uploaded audio file using multipart/form-data.
  • sourceName — Optional label saved with the transcript.
  • options — Optional object. Supported keys:
    • transcriptionModel — Preferred model (e.g., whisper-large-v3); defaults to Horizon’s balanced model.
    • speakerLabels — Boolean; enable speaker diarization (default false).
    • segmentLength — Target characters per chunk (default 800).
    • language — ISO language hint to improve transcription accuracy.
  • webhookUrl — Optional HTTPS URL Horizon should call when the transcription finishes.

​
Sample request

curl https://api.worklet.cloud/v1/extract/audio \
  -H "Content-Type: application/json" \
  -d '{
    "sourceUrl": "https://cdn.example.com/audio/office-hours-42.mp3",
    "sourceName": "Office Hours 42",
    "options": {
      "transcriptionModel": "whisper-large-v3",
      "speakerLabels": true,
      "segmentLength": 700,
      "language": "en"
    }
  }'

# or upload the raw audio (Base64 encoded)

curl https://api.worklet.cloud/v1/extract/audio \
  -H "Content-Type: application/json" \
  -d '{
    "file": "data:audio/mpeg;base64,//uQZAAAAAAAAAAAAAAAA...",
    "sourceName": "Office Hours 42",
    "options": {
      "transcriptionModel": "whisper-large-v3",
      "speakerLabels": true,
      "segmentLength": 700,
      "language": "en"
    }
  }'

​
Response

Returns 202 Accepted with jobId, status, statusUrl, and optional etaSeconds. When processing finishes, result contains transcript chunks with timestamps and, when enabled, speaker labels.

​
Notes

  • Audio longer than 30 minutes always runs asynchronously; use statusUrl or webhooks for completion.
  • Provide language hints for multilingual content to reduce latency.
  • Speaker diarization adds processing time but produces cleaner dialogue segmentation.
  • Poll GET /jobs/{jobId} (equivalent to statusUrl) when you want to check progress or download transcripts later.
  • To upload the audio directly, send multipart/form-data with a file field instead of sourceUrl.

​
x402 flow

Audio extraction is billed per minute via Coinbase’s x402 protocol. A missing proof yields: 
HTTP/1.1 402 Payment Required
Content-Type: application/json

{
  "x402Version": 1,
  "accepts": [
    {
      "scheme": "exact",
      "network": "base-sepolia",
      "maxAmountRequired": "220000",
      "resource": "POST /extract/audio",
      "description": "Horizon audio transcription",
      "mimeType": "application/json",
      "payTo": "0xYourReceivingWallet",
      "maxTimeoutSeconds": 600,
      "asset": "0xYourUSDCContract",
      "extra": {
        "name": "USDC",
        "version": "1"
      }
    }
  ],
  "error": null
}
Send the challenge to your facilitator, run /verify and /settle, then replay the POST with the Base64 payload in X-PAYMENT. Horizon restarts the job and returns X-PAYMENT-RESPONSE on success.