Fix one wrong sentence in a voice-over: 130 characters, not 3,000
Redoing a 130-character sentence costs $0.006175 in Sume TTS plus a $0.01 audio join. Re-synthesizing the 3,000-character script costs $0.1425.

To fix one wrong sentence of 130 characters in a voice-over, regenerate only that sentence and rejoin: $0.006175 for the speech (0.13 x $0.0475) plus $0.01 for a Timeline audio concat, $0.016175 in all. Re-synthesizing the whole 3,000-character script costs $0.1425, nearly nine times as much. The saving is real only if you kept the voice-over as separate takes, so plan that before the first run.
Keep takes, not one file
Split the script by sentence or by paragraph and send each as its own TTS job with the same preset: voice selector, speed, volume, language and format. Store the audio URLs in order. A correction replaces one URL in that list, and one Timeline audio concat job, up to 20 parts, builds the new master. If the script has more than 20 parts, join in two stages.
| Fix | Characters | TTS cost | Join | Total |
|---|---|---|---|---|
| Redo one sentence | 130 | $0.006175 | $0.01 | $0.016175 |
| Redo one paragraph | 600 | $0.0285 | $0.01 | $0.0385 |
| Redo the whole script | 3,000 | $0.1425 | none | $0.1425 |
The seam
Rejoining can leave a short pause or a change in tone at the cut. Use the same preset to keep the voice, and use WAV for the takes so the join adds no padding. If you need exact sentence times, the TTS timestamps option with sentence segmentation gives them; the docs note the boundary_lead_ms default of 70 ms and that emit_audio needs wav or raw.
When to redo it all
A change that affects pacing across the piece, such as switching voice or speed, means all takes must be regenerated, and that is the 3,000-character cost. A fix to a name or a number is the cheap case. The video render also repeats if the voice-over changes under finished video: $0.10 per output minute, rounded up.
A routine for corrections
Number the takes in the order of the script and keep a table of take number, text, audio URL and job id. When the client changes a line, edit its row, send one TTS job, and replace the URL. Rebuild the master with a concat job, check the seam by ear, and rerender the video. Ten corrections at the sentence level cost 10 x $0.016175 = $0.16175, about the cost of one full re-synthesis, but without disturbing the other takes.
If a correction touches the same sentence twice, you pay for it twice, so confirm the text before you send it.
Sources
Related posts
More in Use cases
- Cap a Format run at $4: generation_spend_cap_usd on a before/after ad
Set generation_spend_cap_usd to 4 on a Sume Format run so a before/after ad cannot spend more than $4. What failure looks like and how to read the receipt.
- Four Music 1.0 takes for a 45-second explainer: $0.50 total
Generate four candidate music beds for a 45-second explainer on Sume Music 1.0 at $0.125 flat each, $0.50 total, with a seven-axis brief to narrow the takes.
- Game teaser in 15 seconds: H3 Max start and end frame for $1.50
A 15-second teaser from two key-art frames costs $1.50 on MiniMax H3 Max at 768p on Sume. The image_url and end_image_url body, the 1080p price and the checks.
- Gate announcements: 8 messages in four languages, 32 TTS jobs
8 gate messages of 220 characters in 4 languages is 32 Sume TTS jobs at $0.01045 each, $0.3344 total. Set language and a matching voice per job.
Written by Sume