The screen says 0042, but the voice says “forty-two.” That might be correct for a quantity and wrong for an access code. A speech model needs context. Give it a spoken version that preserves the meaning, then check that version against the original value.

Quick answer: Keep display text and spoken text in separate columns. Identify whether each number is a quantity, date, price, identifier, decimal, or unit. Resolve ambiguous source values first, write the intended reading in words, and listen while following the original.

Download the speech worksheet · Audition a corrected number phrase

Classify the number before rewriting it

A number’s role changes its reading. The year 2026, a room numbered 2026, and a quantity of 2,026 may need different phrasing. Do not run a global replacement that treats them all alike. Preserve leading zeros and signs when they carry meaning.

Display textIntended meaningExample spoken form
0042Four-digit identifierzero zero four two
1.05Decimal valueone point zero five
$12.50US-dollar pricetwelve dollars and fifty cents
5 kgMass in kilogramsfive kilograms
September 25, 2026Explicit calendar dateSeptember twenty-fifth, twenty twenty-six
03/04/2026Locale not establishedResolve the date before writing speech

These examples use English and an explicitly identified US-dollar price. Choose wording appropriate to the actual language and audience. Do not infer a currency or date convention from the symbols alone when the source leaves it unclear.

Resolve ambiguous dates at the source

03/04/2026 could mean March 4 or April 3. A fluent reading can still be wrong. Ask the source owner or consult the authoritative record before normalizing it. Once confirmed, use a written month name in the spoken script.

The W3C’s guidance on clear formatting and punctuation is a useful reminder that presentation can create ambiguity. In narration, the listener cannot inspect a slash or infer the intended format as easily as the writer can.

Keep a value audit beside the narration

Create one row for every important number. Include the original, its meaning, the spoken version, and who or what verified it. Do not replace the visible text with a pronunciation spelling if that would make the screen less readable.

For a range, preserve both endpoints. For a negative value, keep the sign explicit. For decimals, read the digits after the point in the way the task requires. For identifiers, preserve order and leading zeros. A natural-sounding sentence is not enough if any of those details change.

Your reference code is zero zero four two. The package weighs five kilograms. The total is twelve dollars and fifty cents.

Use this only as an audition example. Replace all values with verified project data before producing a real message.

Patch the phrase, then listen against the original

Render the whole sentence containing the corrected value so the voice has surrounding context. Listen once for accuracy while reading the original record, and again for natural delivery within the paragraph. If the phrase changes pace, revise the sentence structure before forcing a word-by-word splice.

ElevenLabs’ text-to-speech best practices discusses text normalization and model-dependent handling of complex text. Preprocessing can help, but the delivered audio still needs verification. A model switch can change a reading that previously worked.

How QuestStudio helps

Use Voice Lab to audition a short corrected passage before rendering the complete script. Keep the same voice and settings during the comparison, and save the accepted spoken version with the source value table.

Use the pause guide if the corrected wording needs breathing room. Use the pronunciation workflow for names and specialized terms, and the natural voice guide for the broader delivery pass.

Frequently asked questions

Why does text to speech read numbers incorrectly?

The model may infer the wrong role or locale. A sequence can represent a quantity, identifier, date, or price, each requiring a different reading.

Should I spell every number out?

Spell out the ambiguous or important values in the spoken script when that improves control. Keep the original display text separately and verify meaning.

How do I make a voice read leading zeros?

For an identifier, write the required digits individually, such as zero zero four two, then listen to the output.

How should I handle an unclear numeric date?

Resolve the intended date from an authoritative source first. Do not guess between month-first and day-first conventions.

Start with one version you can inspect. Audition a corrected number phrase, then use the worksheet to decide what needs another pass.