Video inspect transcribe: no duration hint reserves 1 minute

When you send transcribe: true to Sume video inspect, the transcript is billed at the speech-to-text rate of $0.01 per audio minute. Without a duration_seconds hint Sume reserves 1 minute, and the hint can be at most 600 seconds, so a 10-minute clip reserves 10 x $0.01 = $0.10.

The rule is on the video inspect page, and the live rate is in the catalog. duration_seconds is allowed only together with transcribe: true; otherwise the request returns 400 video_inspect_transcribe_required.

Reserve by hint

The table applies the $0.01 per minute rate to the hint. Whether the final charge matches the reserve is described by the job result, so read it there.

Clips longer than 10 minutes

Inspect accepts sources up to 1800 seconds, but the hint stops at 600. For a 28-minute clip, cut it with video trim first, in ranges of up to 10 minutes, so each pass matches a hint. Three trims at $0.02 plus three transcripts at up to $0.10 is 3 x $0.02 + 3 x $0.10 = $0.36 at the maximum, which is the figure to budget even if the real charge is lower.

{
  "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
  "frames": false,
  "transcribe": true,
  "language_code": "en",
  "duration_seconds": 300
}

Options that come with transcribe

language_code is an STT hint such as en or ko, and auto-detect is used when you omit it. segmentation.mode: "sentence" splits by sentence, with silence_split_seconds from 0.2 to 3. A clip with no audio track returns inspect_source_has_no_audio, so read probe.has_audio first.

• The default mode is sync with a 30 second wait.
• The source clip must be on your workspace's media host.
• language_code, segmentation and duration_seconds without transcribe: true return a 400.

Why the hint matters

The reserve is a hold placed before the work, so a hint that matches the real length keeps the hold small and the estimate honest. With no hint the hold is one minute, which under-reserves a 10-minute clip; with a hint of 600 the hold is $0.10. Send the real length when you know it, for example from the probe in an earlier call.

If you are running many clips, add the figures. Twenty 5-minute clips are 20 x 5 x $0.01 = $1.00 in transcripts. The same twenty without hints would reserve 20 x $0.01 = $0.20 up front, which is why a hint gives a truer picture of what the job needs.

Choosing sentences or raw text

If you need the transcript as sentences with times, set segmentation.mode to sentence. Silence of silence_split_seconds also ends a sentence, between 0.2 and 3 seconds. A smaller value cuts more often, which suits fast speech and short captions; a larger value keeps long thoughts together. Pick the value by reading a sample before you batch.