How to handle numbers, abbreviations and names in text to speech
A date or abbreviation may have several valid readings. Decide what the listener should hear, then make that reading clear in the text instead of expecting the voice to infer your intention.
Write ambiguous numbers as spoken words
Dates, decimal separators and currencies depend on the audience’s language and region. Replace an ambiguous date with a month name. For codes and phone numbers, specify individual digits when necessary. Keep the original formatted version for on-screen text.
‘03/04’ → ‘March fourth’ for a US month/day date, or ‘the third of April’ for a day/month date. ‘$12.50’ → ‘twelve dollars and fifty cents’. Code ‘204’ → ‘two zero four’.
Choose a reading for abbreviations and names
Some acronyms are spoken as words, while others use letter names. Try spaced letters for an initialism, and expand it if the result remains unclear. For a person or brand, confirm the intended pronunciation first. A phonetic respelling can help, but test it in the full sentence rather than assuming it will work.
‘API’ → ‘A P I’, or ‘application programming interface’ on first use. Keep the official name in captions even if the speech script uses a pronunciation aid.
Adapt each language separately
Do not reuse an English respelling in every translation. Letter names, number grouping and stress change across languages. Keep a small list of approved readings for each language and test mixed-language sentences with the chosen voice. Change one troublesome expression at a time. Use a separate narration script so pronunciation fixes do not alter captions or published names.