Numbers in Speech Transcription: Digits or Words?

Last updated: November 6, 2026

Whether to write a number as digits or words is one of the most common questions in transcription guidelines. There is no single right answer, but there is a wrong one: mixing both without a rule.

Two common approaches

• Spoken form (words): the transcript matches exactly what was said. This is typical for speech recognition training data.
• Written form (digits): easier to read, used when the transcript is meant for people.

Some projects ask for both: a verbatim version with words and a normalized one with digits.

Why consistency matters

A model learns the link between sound and text. If the same spoken number appears as digits in one file and words in another, the link becomes noisy. Pick the project's rule and apply it everywhere.

For recording scripts

When you prepare sentences for speakers to read, digits can be read in different ways, so the recordings will not match the expected text. Converting numbers to words before recording avoids this. In Arabic, also decide on dialect: Egyptian or Standard.

Practical steps

1. Read the guideline and note the rule for numbers, dates, times and currency.
2. Convert your sentence list with the numbers-to-words tool.
3. Record the sentences with the voice recorder.
4. Review a sample against the guideline before delivery.

See also how to write numbers as Arabic words.

All articles