Audio detach refuses a 31-minute video: trim it first, $0.43 total

A 31-minute video is 1,860 seconds, over the 1,800 second source limit of Sume audio detach, so the job fails with source_duration_exceeded. Cut the video into shorter pieces with video trim first, detach audio from each piece, then transcribe. For a 31-minute video, four 465-second pieces cost $0.43: $0.08 trimming, $0.04 detaching, $0.31 transcribing.

The limits that shape the plan

Three caps interact. Audio detach accepts a source up to 1,800 seconds and produces at most 900 seconds of audio. The STT request documents a duration_seconds range of 1 to 600, so keep each transcription input at ten minutes or less. A 31-minute file therefore needs at least four pieces to satisfy the ten-minute rule (1,860 / 600 = 3.1, rounded up to 4).

The plan and its cost

Trim takes a start and one of end or duration, and detach can emit 16 kHz mono WAV, the shape the audio detach docs call the STT shape. Trim and detach are $0.02 and $0.01 per job; STT is $0.01 per audio minute (Sume API catalog, read 2026-10-09).

Trim details to get right

Piece starts are 0, 465, 930 and 1,395 seconds, with duration 465 each; the last piece ends at 1,860. Cutting on a silence is kinder to the transcript than cutting mid-word, so move boundaries a few seconds if you know where the pauses are. Trim needs the clip on media.sume.com first, so import it, and every write needs an Idempotency-Key.

Then offset each piece's word times by its start when you merge the transcripts: a word at 12.4 s in the third piece sits at 930 + 12.4 = 942.4 s in the original.

If you only have the audio

If you start from an audio file rather than a video, skip trim and detach: the STT route takes a public HTTPS audio_url and the same ten-minute ceiling applies. The audio detach docs list every refusal code, and the video trim docs cover the range fields.

What the number leaves out

The $0.43 covers only preparation and transcription. It does not include the cost of producing or importing the video, and it assumes each detach and trim bills at its flat per-job price. Catalog reserves are shown in whole cents, so expect a cent of rounding on the STT jobs in practice.

If you want one transcript file at the end, concatenate the four transcripts in order using the offsets above, and keep the piece boundaries in your own records so you can trace any line back to a source time. For a podcast that already exists as an audio file, the same ten-minute rule applies without trim or detach, making the pipeline cheaper: 31 minutes of STT alone is $0.31.