2,400 voice notes of 25 seconds: Sume STT $24.00 direct, $11.20 joined

Transcribing 2,400 voice notes of 25 seconds each (60,000 seconds, 1,000 minutes) on Sume costs $24.00 if you send one request per clip, because each request bills the catalog's 1-cent minimum and a 25-second clip lists at only $0.0042. Joining 20 clips into each 500-second file first and transcribing the 120 joined files costs $11.20: 120 concat jobs at $0.01 plus 1,000 minutes at $0.01.

The five ways

The batch and streaming rows use Microsoft's introductory page prices on 16.667 hours of audio. The direct Sume row is 2,400 x $0.01. The joined row is 120 x $0.01 for the concats plus 1,000 x $0.01 for the minutes, since 120 files of 500 seconds each add up to the same 1,000 minutes.

What the joined route needs

Timeline audio concat takes up to 20 parts and needs Sume-hosted audio, so the clips must be on media.sume.com. The result returns segments[] with each part's start offset, which you subtract from each word time in the transcript to attribute words back to the right clip. That bookkeeping is the cost of saving $12.80. If a clip's speech runs to its very end, words near a boundary need a check against the offsets.

Because a concat of 20 clips of 25 seconds is 500 seconds, each joined file is under the 600-second STT cap, with 100 seconds of headroom.

When not to bother

If you transcribe a few dozen clips a month, direct requests cost cents and the join logic is not worth writing. At 2,400 clips the gap is $12.80 a run, and at Microsoft's batch price the same audio is about $1.67, so the real question for a large archive is whether the transcripts need to be produced inside Sume at all.

Attributing words back to clips

Every part in the concat result has a start offset in segments[]. For a joined file of clips A, B, C the offsets might be 0, 25 and 50 seconds, so a word that the recognizer times at 31.4 seconds belongs to clip B at 6.4 seconds. Build the map once per joined file, then assign each word to the part whose range contains it. The 1-cent minimum is the reason this is worth doing at scale; without it, the direct route would cost $10.00, the same as the joined route's STT line.

Check the arithmetic once more: 2,400 x 25 s = 60,000 s = 1,000 minutes. Direct: 2,400 x $0.01 = $24.00. Joined at 20 per file: 120 files of 500 s, so 120 STT requests at 500/60 x $0.01 = $0.0833 each, $10.00 in total, plus 120 concats at $0.01 = $1.20, which gives $11.20.