Reading web addresses, emails and handles aloud in French

A modern blog post is not made only of sentences. It carries a site address at the end of a paragraph, a contact email in the call to action, a social handle behind an at sign, sometimes a hashtag. A reader glides over these because the brain reconstructs "dot f r" or "at" on its own. A synthetic voice has no such luxury: it must decide what to pronounce, what to spell out, and what to leave silent. This is one of the fastest ways a generic reading gives itself away, because everyone knows what a web address is supposed to sound like and hears a wrong one immediately.

The forms that trip a voice up

Here are the shapes that recur in web copy, the reading you expect, and the mistake you actually hear on a voice that was never prepared for them.

| Written form | Expected reading | Common mistake |
|---|---|---|
| wedispatch.fr | wedispatch dot f r (in French, "point efférre") | reading "dot fr" as one word, or dropping the dot |
| https://wedispatch.fr | wedispatch dot f r | spelling out "h t t p s colon slash slash" in full |
| contact@wedispatch.fr | contact at wedispatch dot f r | reading "at" as the French preposition, or swallowing it |
| @julien | at julien | the at sign dropped, or read as "to julien" |
| #accessibility | hashtag accessibility | the word glued to the sign, or the sign ignored |
| /blog/audio | slash blog slash audio | the path read as one unpronounceable word |
| RGPD (GDPR) | spelled letter by letter | an attempted "French" pronunciation of the string |

Every row is a two-minute test you can run against any tool, ours included. The right-hand column is not hypothetical: these are the readings you hear from voices that predict pronunciation with no model of French web text, where the training data leans heavily English.

Why this is hard for a machine

The core difficulty is that the same character does not read the same way twice. A dot is "point" inside an address and nothing at all at the end of a sentence, where it becomes a pause. A slash is a slash in a URL path, but "sur" in a fraction and part of a date elsewhere. The at sign only really lives inside an address. A neural voice does not reason about these contexts; it guesses a pronunciation from what it saw in training, and French web writing is far less represented there than plain prose.

There is also a matter of taste that is not merely cosmetic. Should the voice read a URL's protocol? Almost never. Nobody says "h t t p s colon" out loud, and a voice that does stretches the sentence with noise the listener has to filter out. Preparing web text well is as much about deciding what to leave silent as what to speak. That work is a set of rules applied before synthesis, not a voice setting, which is the whole argument of the French text to speech page: what matters happens before any sound comes out.

What WeDispatch does about it, without overselling

The common forms in the table are handled by the normalisation applied to every article: a domain name is read with its dots spoken, an email address goes through "at", a URL protocol is not read, and a French phone number is grouped in pairs rather than spelled digit by digit. That is the baseline, treated as a given rather than an option.

What remains are the cases no rule can guess. An unusual top-level domain, a handle that is not a real word, a hashtag in another language: context does not settle these. For those particular readings that must hold across all of your articles, the pronunciation lexicon lets you declare the expected form once, without rewriting your text or regenerating audio already produced. It is the same remedy as for an industry acronym or a place name: one clear rule, written once, that keeps holding.

The writing habit that helps

Half the job is done at the keyboard, and it is a good habit regardless of audio. A cleanly written address, a grouped phone number, a readable hashtag: text that is well kept reads better by ear exactly as it reads better on screen. When you drop a contact address into a post, prefer the full, regular form over a clever abbreviation that neither a human nor a machine parses at a glance.

The rest follows the same principle as acronyms and proper nouns, covered in acronyms in French audio and in pronouncing proper nouns in news audio. A mishandled address, a garbled acronym and a mangled place name are three symptoms of the same gap: a voice never taught to read anything but sentences. The cure is the same, and it fits in one line: good rules before synthesis, and your hand kept on the special cases.

Read next

• The mute e: the French syllable that appears and vanishes
• Teach a voice one word, once, for every article you publish
• Adding an audio version to a WooCommerce store