Week-end, cloud, deadline: how a French voice reads English loanwords
French writing is full of English words. Should a voice say them in English or in French? What a speech engine decides, and how to correct it.
Open any professional blog article written in French and you will find "le week-end", "un manager", "le cloud", "une deadline", "le marketing", "un meeting". These words are French by usage and English by spelling, and they are a real puzzle for text to speech. Should they be pronounced in English, like a native speaker, or in French, the way the person writing them actually says them? There is no single answer, there is a choice, and a good engine has to make the right one, not the most impressive one.
Two symmetrical mistakes
The first mistake is to French-ify everything. An engine that blindly applies French rules to "cloud" reads it as if it rhymed with the French "boue" plus a hard d: nobody says that. The second mistake, more common on models trained mostly in English, is to overdo it: switching to a perfect English accent in the middle of a French sentence, which produces a break as jarring as the opposite. It is the cousin of the problem we describe in why a French synthetic voice sounds English: there, the flaw is a stray accent on French words; here, it is too strong an accent on English words.
The right reading is almost always that of real spoken French: "week-end" is said "week-end" flatly, not with the drawled English diphthong. That is what a journalist, a product lead, a teacher says. The voice should sound like the newsroom, not like an English lesson.
A few cases and the expected reading
| Written word | Expected reading (spoken French) | Common error |
|--------------|----------------------------------|--------------|
| week-end | flat "week-end" | drawled English diphthong, or "vay-ay-kend" (French-ified) |
| cloud | "klaoude" | "kloud" rhyming with French "boue" |
| manager | "manajeur" | "ma-na-gay" (like the French nager) |
| deadline | "dedline" | "day-a-dline" letter by letter |
| marketing | "marketing" with a French g | ending nasalised into a French "-in" |
| live | "laïve" | "live" rhyming with the French "rive" |
| data | "data" with a French a | heavy English "dayta" |
The table shows the real stake: the target is neither pure English nor pasted-on French, it is the pronunciation of usage, the one that goes unnoticed by a French ear. An English word read well inside a French text is a word you did not notice.
What the engine does, and what it cannot guess
On the most common loanwords, the engine applies the French pronunciation of usage rather than a strict English rendering. It is a deliberate stance, consistent with what we argue on the French text to speech page: we tune the tool to read real French, loanwords included, and not to show off a multilingual virtuosity that breaks the flow.
Where it cannot guess is on ambiguous words and brand names. "Live" is said "laïve" (broadcasting live) but sometimes "live" in technical contexts. A product name, a brand, a recent trade term the engine has never met: it will make a default choice, and that choice may not be yours. English acronyms mixed into French add another layer, as we detail for acronyms and abbreviations: is "CEO" spelled out the French way or the English way? That depends on your readership, not on a universal rule.
The move that fixes the reading
When an English word comes up often in your content and its default reading bothers you, you are not stuck with the engine: you open the pronunciation lexicon and you declare the reading you want, once. Your product name, your industry jargon, the loanword your newsroom says its own way: you set it, and it is applied everywhere, without editing each article. It is the same mechanism as for proper nouns, and it is what turns a generic voice into a voice that speaks the way you do.
How to check before you commit
Take a paragraph packed with English loanwords from your own texts and listen to it. Spot each English word and ask yourself: would I say it like that out loud? If the voice flips into a heavy English accent or French-ifies to the point of comedy, you will know in thirty seconds. It is the trial reflex we recommend in testing a voice in ten minutes: the real test of a French voice is not a perfectly French text, it is a French text as people actually write one, with its borrowings and its cheerful inconsistencies.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo