← Blog · 15 November 2026 · Lire en français

Text to speech and wire copy: what happens to datelines, credits and editor's notes in French

French news sites republish a lot of agency copy, and it arrives with its own conventions: a dateline with the agency in brackets, cut marks, editor's notes, an agency sign-off. We ran a typical French wire story through our text preparation to see what is removed, what is rewritten and what reaches the voice untouched.

On a typical French regional news site, a large share of the day's articles started life as agency copy. Some are republished as they came in, some are topped up with a few local lines, and nearly all of them keep the conventions of the wire: a dateline such as "Rennes, 14 nov. 2026 (AFP) -", cut marks written "(...)" or "[…]", an editor's note flagged "Ndlr" (the French equivalent of "Ed."), and a closing credit like "Avec AFP" ("with AFP").

A reader's eye skips all of that. A listener's ear cannot. Every one of these marks becomes a decision for a text-to-speech pipeline: say it, rewrite it, or drop it. Since our speciality is reading French properly, we wanted a factual answer rather than a reassuring one. So we built a sample French wire story using the forms that really appear on news sites and passed it through our rule-based text preparation, the layer that runs on every article before the voice and involves no language model.

Lines removed before reading

Three lines disappeared, by design.

  • "Publié le 14/11/2026 à 09h12" (published on): a line starting with "published on" or "updated on" followed by a digit is removed.
  • "Photo : Marc Le Goff": photo credits, sources, captions and a leading "©" are removed.
  • "Lire aussi : Le budget 2025" (see also): links to other articles are removed, even when written with a full stop instead of a colon.

These rules only apply at the start of a line. A phrase like "une photo : celle de 1998" ("a photo: the one from 1998") inside a sentence is still read, because a rule greedy enough to catch that would eventually eat the journalist's own words. Each audio's preparation log records how many lines were removed, so nothing vanishes without a trace.

What gets rewritten

The dateline is read, and it reaches the voice as "Rennes, quatorze novembre deux mille vingt-six (Agence France-Presse) -". The abbreviated month is expanded, the date is spelled out in words, and the agency acronym is expanded because it sits in our table of French press acronyms and this is its first occurrence. The second mention, "selon l'AFP" ("according to AFP"), stays as an acronym. We expand once, the way a journalist would.

"Mis à jour à 14h32" (updated at 2:32 pm), however, was not removed. The rule expects "updated on" plus a date; with "at" and a time the line slips through, and the voice says the time in full. That is a gap in the rule, not an editorial choice, and we would rather document it than let you discover it by listening.

An upper-case "NDLR" is spelled out letter by letter, following our general rule for unpronounceable acronyms, described in our article on acronyms in French text to speech. A lower-case "Ndlr" is not recognised as an acronym and goes to the voice as written.

What goes through untouched

Several marks reach the voice exactly as they appear.

  • Cut marks: "(...)" and "[…]" are neither removed nor rewritten. What the voice does with them (a pause, silence, nothing) depends on the engine, and we did not measure it by ear on this sample, so we will not claim an outcome.
  • Bracketed additions: in a quote such as "a-t-il dit [à la presse]", the bracketed words are read like the rest. Nothing tells the listener they were added by the newsroom rather than said by the speaker.
  • Agency sign-offs: "Avec AFP" or "(Avec Reuters)" at the end are read.
  • The reporter's byline: "Par Julie Martin, envoyée spéciale à Bruxelles" is read, which is often what you want.

On plans where preparation also goes through a language model, its instructions ask it to remove photo credits, captions, "see also" links and boxes, and say nothing about datelines, editor's notes or agency credits. The outcome on those lines is therefore not guaranteed either way. The rules above always run afterwards, which is why we describe them: they are the only part that gives the same result every time.

Should the dateline be read at all?

This is the real editorial question, and it has no single answer. Radio news says the place ("In Rennes, the regional council...") but not the agency up front, which sounds like a call sign. On a website, naming the agency also has transparency value: the listener knows where the information came from. Two positions hold up.

Keep the place and the source, but at the end. This is closest to radio practice. It can be set per article with the "Text read" button in My audios, which lets you take the dateline out of what the audio reads without touching the published article. A custom text regenerates the audio and is counted as a regeneration. The method is described in the article on choosing what the audio reads.

Keep everything, but have it said properly. This is the simplest option for a newsroom that republishes a lot of wire copy. The pronunciation lexicon does the rest: an entry "ndlr" → "note de la rédaction" applies to the whole word whatever the case ("Ndlr", "NDLR"), before every reading and even in Faithful reading mode. The text shown in the read-along view and subtitles stays the article's own. One limit: the lexicon replaces a word with another, it does not delete it. An empty replacement is ignored, so removing a mention goes through "Text read".

The takeaway

A French wire story read aloud keeps its conventions, except the ones a safe rule can recognise. The date and the agency acronym are read correctly, credits and cross-links disappear, and everything else (cut marks, brackets, notes, sign-offs) is an editorial choice that stays with you. If you are comparing tools, give them the same wire story and look at what they remove without telling you. Add that test to the checklist on our French text to speech page: it quickly shows whether a tool prepares press copy or simply reads it.

Give your articles a voice with WeDispatch

This blog is itself voiced by WeDispatch. Curious how it sounds on your content?

Book a demo

Read next