45-minute recording: audio detach fails past 1800 s, what to do

A 45-minute video is 2700 seconds, and Sume audio detach rejects any source longer than 1800 seconds with source_duration_exceeded. It also returns at most 900 seconds of audio per job. So a 45-minute recording cannot go through detach whole; you have to cut it into pieces of 1800 seconds or less on the video side, then detach each piece with ranges of 900 seconds or less.

The arithmetic

Source limit is 1800 s, which is 30 minutes. Output limit is 900 s, which is 15 minutes. For 2700 s you need at least 2 sources (2700 / 1800 = 1.5, rounded up to 2) and at least 3 detach ranges if each is 900 s (2700 / 900 = 3). The Sume docs price each detach job at $0.01, so a plain 3-range plan across 2 sources is 3 jobs, or $0.03. Trim jobs are separate and priced on their own docs page.

Cutting the source first

Use video trim to make a new MP4 cut for part A and part B, or use the range field on part A directly if you only need audio from a window. The detach docs say a range with an end at or before start, or a window longer than 900 s, fails with audio_detach_range_empty. A range.start past the source duration fails with detach_start_past_source.

Then transcribe or join

Each detached piece can go to speech-to-text. Sume's own STT line is $0.01 per audio minute, with a per-call hint maximum of 600 s on video inspect, so a 45-minute recording is 45 x $0.01 = $0.45 of STT line cost, before the inspect's compute reservation. To stitch audio pieces back together without gaps, use timeline audio concat, up to 20 parts, as wav.

A related post covers a 12-minute interview done as two STT jobs; the same logic scales here.

• Source over 1800 s: source_duration_exceeded.
• Window over 900 s: audio_detach_range_empty.
• No audio track: detach_source_has_no_audio.