When narration reads a phone number as a large quantity, revise the spoken script rather than changing the visible number. A screen label and a voice instruction serve different purposes. Microsoft documents pronunciation and number-related speech controls in its SSML pronunciation reference. Those controls apply to supported speech systems; do not assume an arbitrary video generator accepts SSML.
Give each number a meaning
Make a two-column script: approved display text on the left, intended spoken wording on the right. Label each numeric field as a telephone number, calendar date, year, price, quantity or identifier. A digit sequence can be correct yet require a different spoken form depending on its purpose.
Ask the content owner to confirm ambiguous dates and units. Do not infer whether a slash-separated date uses month-first or day-first order. Keep contact numbers in a restricted working document when they are private, and use invented numbers in public examples or demonstrations.
Render a small speech test
For a hypothetical event, the visible arrival time might be “18:30” while the spoken script says the equivalent of “six thirty in the evening”. A contact number would normally be read as digits, while an attendance count would be read as a quantity. These are assumed examples, not real event details.
If the chosen speech service supports explicit interpretation controls, apply them only to the relevant fragment and test the result. If it accepts only plain text, write the intended spoken words directly. Keep that pronunciation-friendly script separate from captions and on-screen graphics so a phone number is not accidentally displayed as a long verbal sentence.
Listen to the fragment without watching the screen. Write down the number you heard and compare it with the approved information. Then watch with the screen visible and check whether the voice and display agree. Test the same voice and settings intended for the final production; a different preview voice is not a sufficient proof.
Approve before rebuilding the film
Keep the approved audio fragment and script version. Replace the affected narration only after it passes, then inspect any changed timing around the edit. Longer spoken wording can require more screen time; do not cut off the final digit merely to preserve an old shot length.
Acceptance means a listener can recover the intended information and the visible details remain exact. This does not verify whether the underlying event date or contact number is true. For a project in Imagild Film Studio, settle the spoken information before assembling the final narration and captions.