Translating into French for audio: the English formats a French voice will read literally

If you publish in French but write, source or brief in English, a lot of your French content started life as English: a global press release, a product page, a research summary, a localized blog post. Modern translation, human or machine, usually produces sound French sentences. What slips through are the formats: how a number is punctuated, how a date is ordered, which units are used. A French reader fixes these silently with their eyes. A French listener cannot, because by the time they hear it the voice has already committed to a reading.

This is not a translation style guide. It is a test. We ran typical sentences from translated content through the module that prepares French text before speech synthesis at WeDispatch, the step that turns digits, dates and symbols into words, and recorded exactly what it hands to the voice. Anyone with an account can reproduce every line in the reading preview, which costs no credits.

The problem is the convention, not the language

A French text preparation applies French conventions, which is exactly its job. In French, the comma marks decimals and a space separates thousands. So "1,200 dollars", written the American way, means "one point two hundred dollars" to a system that reads French correctly. The system is not wrong. The text is written in a different convention from the language it claims to be in.

That is what makes these errors hard to catch in review. The text looks French, the voice sounds French, and the figure the listener hears is off by a factor of a thousand.

What actually reaches the voice

| Written in the translated text | What the preparation sends to the voice (in French) | What the translation should have said |
|---|---|---|
| 1,200 dollars | "un virgule deux cents dollars" (1.200) | 1 200 dollars |
| 2.5 billion dollars | "deux virgule cinq billion dollars" | 2,5 milliards de dollars |
| €9.99 | "neuf euros.99" (the cents are left hanging) | 9,99 € |
| 4.5/5 | "quatre point cinq slash cinq" | 4,5 sur 5 |
| 10/11/2026 (meant as October 11) | "dix novembre deux mille vingt-six" | 11 octobre 2026 |
| 11:30 a.m. | "onze heures trente a.m." | 11 h 30 |
| 3 PM | "trois pé emme" (spelled out) | 15 h |
| 150 lbs | "cent cinquante lbs" | 68 kg |
| 60 mph | "soixante mph" | 97 km/h |
| Mr. Smith | left unchanged | M. Smith |
| 4th of July | left unchanged | 4 juillet |

In fairness, plenty is handled without any help. "3.5%" becomes "three point five percent" in French, "9.99 €", with the symbol after the number, is read as euros and cents, "$5" becomes "five dollars", "75°F" becomes "seventy-five degrees Fahrenheit", "vs." becomes "contre" and "Q3" becomes "third quarter". In most cases where the English format is unambiguous, the preparation reads it the French way. A price written "€9.99", symbol first with a decimal point, is the exception, hence its row in the table.

Why it does not guess

You might want software that recognizes "1,200" as an American thousand. Our preparation follows the opposite rule: when in doubt, leave it alone. "1,200" can be an English-style thousand, or a perfectly French decimal with three digits after the comma: a distance in kilometres, a rate, a betting price. A rule that "fixed" the first case would silently break the second, and nobody would know where the error came from. A mistake that is visible in the text is better than one the tool invented.

The date is the same story. In France, 10/11 means 10 November, and that is how it is read. The software cannot know the sentence came from a New York press release. You can.

Three families of errors to look for

Numbers. Swapped decimal points and commas, thousands separated by commas, and above all the scale false friend: an American billion is a French milliard, while a French billion is a million million. Read faithfully, "billion" makes the listener hear an amount a thousand times too large, and unlike a reader they have no way to double-check. Our guide to reading numbers aloud in French lists the forms a French ear expects.

Dates and times. Month-first dates are the most dangerous format because they produce a valid, wrong date: 3/4 becomes 3 April instead of 4 March, and nothing sounds off. Writing the month as a word removes the problem for good. "a.m." and "PM" are not French abbreviations, so they are treated as acronyms and spelled out letter by letter.

Units and titles. Miles, pounds, feet and Fahrenheit are read correctly, but the number means nothing to most French-speaking listeners. That is a localization issue more than a pronunciation one, and audio makes it louder: a reader can pause to convert, a listener has to do the maths while the sentence keeps going. "Mr." stays as written; only "Dr.", close enough to French usage, is expanded.

A workflow that catches them

None of this needs a developer:

• add a formats pass to the review of translated copy: search for commas inside numbers, slashes inside dates, and the strings "a.m.", "PM", "billion", "mph" and "lbs". It takes two minutes in any editor;
• open the reading preview before voicing an important piece: it lists every transformation, before and after, without synthesis and without spending credits. "Un virgule deux cents" is hard to miss there;
• fix the text, not the audio: the correct form helps readers, search engines and subtitles too. The pronunciation lexicon is for names and in-house acronyms, not for patching a wrong format.

This is one concrete case of the principle behind our approach to French text to speech: good audio starts with text that is written in the language being read, conventions included. If translated content is a large share of what you publish, our notes on writing for the ear are worth a second look.

Read next

• Quote, unquote: should a text-to-speech article announce its quotations?
• Changing your audio volume mid-subscription: what happens to your credits and your invoice
• French text-to-speech for children: choosing a voice for young listeners