Quote, unquote: should a text-to-speech article announce its quotations?

A reader sees the quotation marks before the first word of the quote. In one glance they know where the reported speech begins and ends, and that what sits between the marks is not the newsroom's own claim. A listener gets none of that. French guillemets (« ») are silent, and a synthetic voice reading "We will not close a single class" does not, on its own, tell you whether the reporter is asserting it or quoting the education officer.

Spoken language solved this long ago, and the solutions carry over to text-to-speech. Here is what they are, what each costs the listener, and what our "Announce quotes" setting actually does.

How broadcasters handle it without any tool

A radio journalist reading out a statement has three techniques and blends them without thinking.

• Attribution first. "The education officer was clear: we will not close a single class." Speech names the speaker before the words, whereas print, in French as in English, often puts the attribution after the quote. This is the most natural technique and adds no words at all.
• A framing formula. "Quote... end quote", "in her own words", "and I'm quoting here". French radio says « je cite... fin de citation ». It is used when the boundary really matters: a disputed claim, a word the newsroom does not endorse, a court ruling.
• The voice itself. A slightly different tone or pace. Synthesis can approximate this by giving quotations a second voice, which we cover in using multiple voices for quotes.

The framing formula is the most explicit of the three, and the heaviest. It adds a few syllables before and after every quotation. On a news brief with two quotes, nobody minds. On an interview with fifteen, it turns into a verbal tic.

When announcing actually helps

Three situations justify it.

The article is read by a single voice. With no second voice, nothing distinguishes reporter from source to the ear. The announcement is the only marker, and it is a real service, especially to a blind or partially sighted listener who has no text to glance at.

The quote puts someone on record. An accusation, a controversial remark, an extract from a judgment: the listener must know beyond doubt that the newsroom is reporting, not asserting. That is why media lawyers like to see quotation marks, and the announcement is their audio counterpart.

The quotes are long. Beyond two sentences, even with a second voice, listeners can lose track of who is talking. An "end quote" gives them the frame back.

Conversely, it weighs down a profile or a Q&A where the quoted speech is the substance of the piece. There, a second voice, or attribution placed before the quote, does the job.

How the tool decides what counts as a quotation

An announcement is only useful in the right place. "Quote" spoken before a three-word phrase sounds like a glitch. Detection is therefore deliberately cautious, and these are the rules as written in the code:

• the passage sits inside French guillemets (« ») or curly quotes (“ ”); straight quotes only count right after a colon, because they open and close with the same character;
• it starts its own sentence: what comes before is a full stop, a question or exclamation mark, an ellipsis, a colon, a semicolon, or nothing;
• it runs to at least 80 characters, excluding the marks, roughly the length of one spoken sentence.

So when a mayor's statement is reported as calling a budget a « chèque en blanc » mid-sentence, nothing is announced: that is a phrase inside the reporter's own sentence, and it is read as such.

What the setting does, precisely

On the Voice screen, the "Announce quotes" card has two fields, "Before" and "After". You type your own words: « Citation » and « Fin de citation » for a French site, "Quote" and "End of quote" for an English one, or just one of the two. Left empty, nothing is said.

The fields have guard rails. Forty characters at most on each side, because this is a formula, not a sentence. Letters, spaces, apostrophes, hyphens and commas only: a digit or symbol would be read oddly. Trailing punctuation is stripped and a full stop is added so the voice pauses.

The announcement is spoken, not written. It is read by the article's main voice, even when the quote has a voice of its own. It does not appear in the displayed text, in the word-by-word highlighting, or in subtitles: the published text stays the journalist's. The audio's provenance record does note how many announcements were spoken and that they are absent from the displayed text, so the newsroom can see what was added in the audio.

The setting works with a single voice: on its own, it is enough to switch quotation detection on. And a quote you choose to leave on the article voice, from an audio's detail page, is still announced.

Where it still stumbles

Take the sample sentence from our Voice screen, in its French original, where the attribution follows the quote:

« Nous ne construisons pas un quai, nous construisons la capacité d'assembler des flotteurs de mille tonnes », a résumé la directrice du port.

("We are not building a quay, we are building the capacity to assemble thousand-tonne floaters," the port director summed up.)

With announcements on, the listener hears: "Citation. [the quote]. Fin de citation. A résumé la directrice du port."

The attribution arrives after the closing formula and sounds stranded. That is the honest limit of the technique: the written pattern (quote, then "she said") is not a spoken pattern, and no announcement converts it. The fix is editorial and free: when a piece is meant to be heard, put the attribution before the quote ("The port director summed it up: ..."). More advice of that kind is gathered in writing for the ear.

In short

Announcing quotations is neither a luxury nor a given. It is an editorial choice, made per publication. With one voice and quotes that put people on record, it is almost always worth it. With a second voice and many quotes, you are usually better off without it and with careful attribution. Either way, a tool should only announce what really is reported speech. That standard for reading French, more than any count of settings, is what drives the work described on our French text-to-speech page.

Read next

• 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
• When your article audio service goes down: what readers actually see