Sume STT reserves 1 cent by default: send duration_seconds (1-600)

Sume's speech-to-text price is $0.01 per audio minute, and the admit-time estimate assumes one minute if you leave out duration_seconds. A job can be at most 10 minutes, which is 600 seconds and a 10-cent maximum. Send duration_seconds as an integer from 1 to 600 so the estimate reflects your real file.

The numbers in the catalog

The sume/stt-1.0 rate card lists unit per audio minute, price $0.01, an estimate of 1 cent, a minimum of 1 cent and a maximum of 10 cents. The basis line spells it out: absent duration is estimated as one minute, the minimum is one second and the maximum is ten minutes.

Why it matters for a bulk run

A wallet or a spend cap that checks the estimate before a job starts will treat an unlabelled file as one minute. If your files are really nine minutes long, a run of fifty files looks like 50 cents at admission and ends near $4.50. Sending the real duration keeps the preview honest. You can read the length with ffprobe before you submit, which is cheap and free.

• Measure duration locally and send it as whole seconds.
• Round up, not down, if your tool gives fractions.
• Never send more than 600 seconds; split long audio into chunks.

Example: fifty nine-minute files

Say you have fifty interview files, each nine minutes long. Omitting duration_seconds makes each look like a one-minute, one-cent job at admission, so the preview shows about 50 cents. Sending 540 for each makes the preview about $4.50, which matches the list rate of $0.01 for each of 450 minutes.

If your spend cap is close to the real cost, the second number is the one that keeps the run from stopping halfway.

What it does not cover

The catalog gives the per-minute list price and the bounds. It does not say how the final charge rounds sub-minute audio, so check a real job's receipt before you build a precise bill. The audio endpoint map linked below shows which route to call for what.