Voice and audio
Making Text to Speech Say Numbers, Dates and Acronyms Correctly
Generated voices read exactly what is written, which is the problem with 2026, NASA and 3/4. How to write a script so the voice says what you meant.
3 min read · Published 12 October 2026
A text to speech voice does not know what you meant. It reads what is on the page and decides, from the characters alone, how to say them. Most of the time that works. The trouble starts with the three kinds of text where the written form and the spoken form are not the same thing: numbers, dates and abbreviations. A voice that stumbles on those sounds robotic in exactly the places where a listener is paying attention.
The fix is almost never in the voice settings. It is in the script. You are writing for a reader who will follow your spelling to the letter, so you spell things the way you want them said. A few minutes of editing before you render saves several renders afterwards.
Numbers
Digits are ambiguous because English reads them in several ways. The figure 1999 might be said as nineteen ninety nine or as one thousand nine hundred and ninety nine, depending on whether it is a year or a quantity. The engine guesses from context, and the guess is sometimes wrong. The safest approach is to write the number the way you want to hear it.
| Written | Risk | Write instead |
|---|---|---|
| 1999 | Year or quantity | nineteen ninety nine, or one thousand nine hundred ninety nine |
| 3.5 | Might read as a date | three point five |
| $4.99 | Currency order | four dollars ninety nine |
| 50% | Usually fine | fifty percent, if it stumbles |
| 1st | Ordinal | first |
Dates and times
Dates are the worst offender, because the same characters mean different things in different countries. The string 04/05 is the fourth of May in one place and the fifth of April in another, and a voice has to pick one. Write dates in words: the fifth of April, or April fifth. The same goes for times. Write half past three or three thirty in the afternoon, not 3:30pm, unless you have already checked how the voice handles it.
Acronyms and abbreviations
Some acronyms are said as words, such as NASA and radar, and some are spelled out letter by letter, such as FBI and USB. A voice usually gets the common ones right and the unusual ones wrong. If an acronym is wrong, spell it the way it sounds. For letters, put spaces between them, as in U S B. For a word, write it phonetically, as in nasa, with the case that works.
- Spell out any abbreviation whose spoken form is a whole word, such as doctor for Dr.
- Put spaces between letters you want read individually.
- Write unusual acronyms the way they sound, then listen to check.
- Avoid symbols such as ampersands and slashes; write the word instead.
Names and unusual words
A name the voice has never seen is guessed from its spelling, and spelling is a poor guide in English. The fix is the same as for acronyms: write it phonetically in the script, even if the result looks odd on the page. Nobody will see the script, only hear the audio, and a name said correctly is worth more than a name spelled correctly.
Keep a short list of the respellings that work for your subject, such as the names, places and terms that come up repeatedly. Pasting from the list takes seconds, and the result is consistent across every video.
Test the awkward parts first
You do not need to render a whole script to find out whether a number sounds right. Paste the awkward sentences on their own into the voiceover tool, listen once and fix what is wrong. A short test costs a few seconds and can save you from rendering a ten minute script, discovering a mistake at minute eight and starting again.
The goal is a script that reads aloud naturally to a person, because a voice is a very literal person. If you would stumble reading a line out loud, so will it. Edit until you would not.
Common questions
Why does text to speech read numbers wrongly?
Because digits are ambiguous. The same characters can be a year, a quantity or a code, and the engine has to guess. Writing the number the way you want it said removes the guess.
How do I make a voice say an acronym letter by letter?
Put spaces between the letters, as in U S B, so each is read on its own. For an acronym that is spoken as a word, write it phonetically instead.
Should I write dates in numbers or words?
In words. A numeric date means different things in different countries, and a voice has to pick one. Writing the fifth of April leaves no room for error.
How do I fix a mispronounced name?
Spell it the way it sounds in the script. The audio is all anyone hears, so a phonetic spelling that sounds right beats a correct spelling that sounds wrong.