100 short TTS lines: one job or 100 jobs? The 1-cent floor

If you send 100 separate TTS jobs of 40 characters each on Sume, you pay about 100 cents, because every job is rounded up to at least one cent. The same text as one 4,000-character job costs 19 cents. Whether you can use the single job depends on whether you need 100 files or one.

Where the floor bites

Sume bills TTS per character, but each job has a one-cent minimum and is rounded up to whole cents. A 40-character line is $0.0019 of speech and is billed as 1 cent. A job of 4,000 characters is $0.19 exactly.

Getting 100 clips from one job

The TTS request accepts a segmentation setting with mode sentence. It requires word timestamps to be turned on, and it has an emit_audio option that defaults to true. That lets one synthesis job describe sentence-level boundaries, so you can cut the output into lines rather than paying per line. The sentence segmentation post linked below covers boundary_lead_ms.

{
  "voice": {"id": "<voice uuid>"},
  "transcript": "Line one. Line two. Line three.",
  "timestamps": {"words": true},
  "segmentation": {"mode": "sentence"}
}

Checking your own numbers

Add up the characters before you decide. Take the line count, multiply by the average length, and multiply by $0.0000475 for the one-job figure. For separate jobs, charge each line the larger of one cent or its own rounded-up cost. Lines under 211 characters all land on the floor of a cent or a few cents, so very short lines are where batching saves the most in percentage terms.

If the lines are things like button labels or notification sounds, the absolute amounts are small either way: a hundred separate jobs is a dollar. Optimise for how you want the files delivered, and treat the savings as a bonus.

The tradeoff

One job gives one take. If you want to regenerate line 37 alone, you either re-run the whole script or send that line as its own one-cent job. A practical split is one job for the bulk and individual jobs for retakes. Lines that really are independent, like unrelated prompts in different voices, belong in separate jobs anyway, since a job carries one voice.