One TTS job with sentence segments, or one job per sentence?

Use one TTS job with sentence segmentation. On Sume each job is billed per character and then rounded up to a whole cent, so a 1,350-character script costs 7 cents as one job, 12 cents as twelve sentence-sized jobs, and segmentation returns the sentence boundaries inside the single job.

How the rounding works

The TTS price is $0.0475 per 1,000 characters, billed per job, rounded up to the cent, with a one-cent minimum and a 95-cent maximum at 20,000 characters. Characters are counted as the JavaScript string length, so spaces and punctuation count. The consequence is that splitting a script into many small jobs pays the rounding many times.

What segmentation gives you

Send timestamps: { words: true } and segmentation: { mode: "sentence" } on one job and the result carries gapless sentence segments[], where each segment ends exactly where the next begins. The default boundary_lead_ms is 70, meaning the cut falls 70 ms after a sentence's last word and the next segment absorbs the pause. Segmentation requires word timestamps.

That is enough to place a picture cut on each sentence or to build caption lines, with no extra jobs. If you need the audio itself sliced, set emit_audio with a wav or raw container. The slices do not come from an mp3, which is another reason to choose wav for production audio.

curl -X POST https://api.sume.com/v1/tts-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: seg-demo-001" \
  -d '{
    "transcript": "Meet the new kettle. It boils in ninety seconds. Order today.",
    "voice": { "id": "VOICE_ID" },
    "output_format": { "container": "wav" },
    "timestamps": { "words": true },
    "segmentation": { "mode": "sentence" }
  }'

When separate jobs are right

Otherwise take the single job. The five-cent gap in the table is small once, and it adds up across a catalogue: at 200 reads a month the twelve-job habit costs $10 more than one job each.

• Different moods per sentence. A job has one emotion guide, one speed and one volume.
• Cheap retakes. Redoing one 120-character line is 1 cent, while redoing a 1,350-character read is 7.
• Parallel runs. Separate jobs can run at once, which matters if a script is near the 20,000-character limit.

Check your own count

Do not trust the 15 characters per second rule of thumb for your voice. Measure it from the word timings of a real job: characters divided by the last word's end time. Then your estimate for the next script is a multiplication, and your sentence budget is exact.

A worked month

A team publishing 40 product explainers a month, each about 1,350 characters, pays 40 times 7 cents, which is $2.80, with one job each. If the same team renders every sentence as its own job, with twelve sentences per explainer, the bill is 40 times 12 cents, $4.80. Neither number is large, and the argument for the single job is not the money. It is that one job gives you one audio file with one consistent take, where twelve jobs give you twelve takes whose levels and pace you must match later.

Consistency is the quiet benefit. A voice reading one script in one pass keeps its pacing across sentences. Twelve independent calls start each sentence fresh, so a join may sound like twelve people reading in a row.

A rule for the team

Write it down: one job per script unless the mood changes, segments for the sentence boundaries, wav for anything that gets sliced. A rule of three lines settles the argument for every later script and keeps the receipts predictable.