What Is Text Normalization in Text to Speech?
Text normalization is the step in text to speech that converts things that are not plain words, such as numbers, dates, currencies and abbreviations, into the words that should actually be spoken.
Why it is needed
A voice cannot say "3/4" or "Dr." or "$2.50" without deciding what they mean. Is 3/4 a fraction or a date? Is "St." Saint or Street? Normalization makes those choices, and it can guess wrong, especially across languages and regions.
Things that trip it up
- Dates and times, which are written differently around the world.
- Currencies and units.
- Acronyms that could be read as a word or as letters.
- Numbers in phone numbers, addresses and product codes.
- Abbreviations with more than one meaning.
How to avoid surprises
Write the text the way you want it to be said. Spell out numbers, dates and units, such as "two dollars and fifty cents" or "the third of March". Separate the letters of an acronym with spaces if each should be said. Then listen, and add a pronunciation rule for anything still wrong.
Normalization and GetMeAudio
The cloud voices normalize text themselves, and the results vary by voice and language. A free draft is a quick way to catch obvious problems, and the script cleaner tool tidies text pasted from PDFs before you voice it.
Frequently Asked Questions
Why does the voice read a number strangely?
It had to guess whether the number was a quantity, a year, a phone number or something else. Spelling it out removes the guess.
Does normalization change my text?
It changes only what is spoken, not what you typed.
Related
Unlimited free drafts · full-quality previews · pay only for your final downloads.
Open the editor →