Business terms pronounced right: Decagon tuning vs fixing a TTS script

Decagon says Voice 3 lets a business customize voice selection, pronunciation of business terms and delivery. With a script-to-audio job you do the equivalent in the text: respell the term, test one sentence, then lock the wording and generate the full read.

Where the control lives

The two products place the control in different spots.

A test loop for one term

Names, drug names, abbreviations and acronyms are the usual offenders.

• Pull the 5 terms most likely to be wrong from the script.
• Put each in a short sentence and generate five one-sentence jobs.
• Listen and mark each pass or fail.
• For a fail, respell it phonetically in the text, or split an acronym into letters.
• Regenerate only the failing sentences and keep the passing ones.

What it costs on Sume

Sume charges per character at $0.0475 per 1,000 with a one-cent minimum per job, so five 80-character test sentences cost 5 cents. Because each result has sentence slices and word timings, you can replace one sentence in a finished narration and rejoin with timeline audio concat for $0.01 per job instead of regenerating everything.

Keep the job id of the approved take, since the completed job records the voice, language and settings used.

Write a pronunciation sheet

Put the final respellings in a shared sheet next to the script: term, respelling, approved job id. The next voiceover then starts from the sheet, and a new editor does not repeat the same listening round. Also add the term to captions with script_text, so the burned text shows the real spelling while the audio uses the respelled one.