Wedding toast transcript: video inspect with sentence segments

To get a written keepsake of a wedding toast, call POST /v1/video-inspect with transcribe: true on the recorded video, and add segmentation: { "mode": "sentence" }. The response carries the transcript text, words[] with timings and gapless sentence segments[]. The public rate is $0.01 per audio minute, so a 4-minute toast is about $0.04 plus the inspect's own Modal compute.

Source: Video inspect, read 2026-10-04.

Request

Send frames: false to skip the stills when you only want text. Everything that tunes the transcript is nested under transcribe.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: toast-transcript-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/toast.mp4",
    "frames": false,
    "transcribe": true,
    "language_code": "en",
    "duration_seconds": 240,
    "segmentation": { "mode": "sentence" }
  }\

Fields that matter

Set duration_seconds as a hint to size the hold: if you omit it, Sume reserves one minute, and the maximum hint is 600 seconds. The charge never exceeds the hold.

Failure cases

If the video has no audio you get inspect_source_has_no_audio. Probe with frames: false and no transcript first, and read probe.has_audio. Any of language_code, segmentation or duration_seconds without transcribe: true is a 400 video_inspect_transcribe_required.

Proofread before printing

Speech-to-text will mishear names and in-jokes. Read the text against the video before you print it, and fix the names by hand. If you want the toast burned in as captions instead, POST /v1/video-captions takes a script_text that aligns to the speech.