France, Belgium, Switzerland, Quebec, Africa: one tool for every variant of French
French is not read the same everywhere: numbers, regional words, accents. What a voice tool built for French actually adjusts from one variant to the next, and what is still out of its reach today, stated plainly.
French is spoken across four continents, and it is not said the same way in each. A single voice tool therefore has to read a Paris text as well as a Brussels, Geneva, Montreal or Dakar one. The real question is not whether it can, it will read the words in every case, but what it truly adjusts from one variant to the next, and where it stops. Here is the honest inventory: what the tool sets today, and what remains out of reach.
What actually changes from one variant to the next
Three things separate French from one country to another, and they do not pose the same problem to a machine.
Vocabulary first. Magasiner and courriel in Quebec, septante and nonante in Belgium and Switzerland, dépanneur, poutine, drache: these are French words, with French spelling. A synthetic voice reads them without flinching, because it reads letters, not a regional dictionary. Local vocabulary is not, in itself, an obstacle.
Numbers next, and this is where it gets tricky. A Belgian or Swiss text writes "septante" and "nonante" where France writes "soixante-dix" and "quatre-vingt-dix". If the number is spelled out in letters, the tool reads it as written, no problem. But if the text writes "70" or "90" in digits, the tool voices them by the France system, so "soixante-dix" and "quatre-vingt-dix", even for a Belgian reader expecting "septante" and "nonante". That is a genuine mismatch, and it is worth knowing.
Accent last, meaning prosody, vowel colour, the melody of the sentence. It is what makes an ear recognise "someone from here" or "someone from France". And it is the heaviest dimension to change, because it does not live in a word, it lives in the voice itself.
What the tool adjusts today
Two levers are real and available.
The first is targeted pronunciation correction. It lets you set, once and for all, how a specific word must be read, and the tool then respects it everywhere. It is the lever that fixes regional proper nouns, place names, tricky words. A Swiss or Quebec editor pins their recurring vocabulary upstream this way, and never returns to it.
The second is the choice of default voice, including per language on a multilingual site, as we detail for multilingual Belgian sites. You choose the voice that will carry your content, its timbre, its pace, its gender.
What these two levers do not cover is automatic conversion of digit-form numbers to the regional variant. Today, "90" reads as "quatre-vingt-dix". If you publish for a Belgian or Swiss readership and the "nonante" reading matters to you, the workaround exists but it is manual: spell the number out in the text. It is not elegant, it is effective, and we would rather tell you than hide it.
What remains out of reach
Accent does not adjust. Our French voices carry the standard accent of France. We do not offer, today, a voice with a Quebec, Belgian, Swiss or West African accent. A Quebec text given to the tool is read correctly and stays fully intelligible, but it will sound "from France", and we showed this in detail in our article on Quebec French. The same holds for French-speaking Switzerland, which we cover here: intelligibility is a given, the local accent is not.
For many uses, trade press, institutional content, documentation, professional blogs, that intelligibility is more than enough, and accent barely matters. For a newsroom that makes spoken proximity a brand argument, it is a real limit, and a decision criterion to weigh before testing, not a disappointment to discover after.
Our stance
One tool for every variant of French means one clear thing: it reads correctly in all of them, and it gives you the settings to correct the words. It does not mean it imitates every accent, because we do not build a half-solution that would ring false everywhere. That is the line our French text-to-speech page defends: better to read one language well and say where the tuning stops, than to promise five accents and hold none. The variant, we adjust where it is honest to; the accent, we do not fake.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo