Checklist
Check speech controls with a listening rubric
Separate a text-to-speech system’s advertised controls from the pronunciation, meaning and timing a listener actually receives.
THE SHORT ANSWER
A control being available and its output meeting a listening requirement are separate observations.
Start with a bounded claim
Soniox’s archived launch post advertises expressive controls and streaming speech. Its current product page also describes audio tags for delivery and playback while text arrives. These are publisher descriptions. The page’s demonstrations were not listened to here, and this resource does not repeat its pricing, language totals or quality superlatives as independently verified facts.
Write a small listening script
Use an original, harmless script containing an abbreviation, a name, a number and a sentence whose meaning depends on emphasis. For each item, define the intended pronunciation or meaning before listening. Compare an unmodified reading with one changed control. Ask a reviewer to mark intelligibility, omitted words, unintended emotion and whether an interruption leaves the listener with a misleading partial statement.
Keep engineering and taste apart
Record actual audio duration and the conditions under which playback starts, rather than using a subjective label such as fast. Keep preference notes alongside intelligibility and content checks. Use only authorized voices and material in any later experiment. No speech was generated, cloned or evaluated for this checklist. It is an evaluation design aid; voice performance and finished audio production require their own review.
Put it to work.
- Define intended pronunciation.
- Change one control.
- Check omissions and emphasis.
- Record timing separately from preference.
Go to the source.
Original material, with its own scope and publication date. This note is an explanation, not a CooScope performance test.
Source published · Editorial review
Archived English post text; primary sources reviewed separately. Generated annotations, media and missing replies are outside the evidence scope.
- @soniox_ai — archived X post (2026-08-11)
- Soniox — Text-to-speech capability descriptionsLink / source review: 2026-10-15