The reading sheet: check how a French article will be read before you pay for the audio
With most text-to-speech tools, you find a misreading by listening to audio you have already paid for. WeDispatch's reading sheet shows beforehand what will be said differently, what is uncertain and what we cannot read. How it works on a real paragraph, and where it stops.
Here is the usual loop with a text-to-speech tool. You paste an article, start the generation, pay for the characters, then listen. Forty seconds in, an agency acronym comes out wrong, or a word with two possible readings goes the wrong way. You fix it, regenerate, pay again. Next morning, on the next article, the same mistake is back.
The mistake itself is only half of the cost. The other half is when you find it: after the spend, by ear, one full listen at a time. The reading sheet moves that moment earlier. Before anything is generated, it shows what the tool is going to do with your text.
This matters most for French, which is the language we focus on. Dates, amounts, ordinals and acronyms are written one way and said another, and a French newsroom writes a lot of them.
Three blocks, in a deliberate order
You paste a text into the Pronunciation screen of your account, click "See how it will be read", and the answer comes back instantly in three parts.
What will be said differently. Every passage our French reading rules will transform, with the before and the after: a date, a time, an amount, an ordinal, an acronym expanded or spelled out. These changes will happen. This is the certain part.
What you should check. Words spelled the same and said differently depending on meaning. In French, "fils" can be a son or electrical wires; "est" can be the verb or the East. The sheet flags them with both readings and does not decide for you, because correcting the wrong way would introduce an error the voice might not have made. We go into this in our piece on French homographs.
Acronyms we cannot read. The ones we can neither expand nor spell, which will reach the voice as written. They are the obvious candidates for your pronunciation lexicon.
Certainties first, doubts second, action last. A screen that opened with its doubts would look like a product that hesitates.
A real paragraph, unedited
We ran a short French town council report through the sheet, the kind of paragraph a local news site publishes every week. It contains a date written "12/03", an amount written "1,2 M€", a percentage, a year, the abbreviation "M." for Monsieur, two acronyms, a time written "18 h 30", an ordinal and the word "fils".
The summary line read: "9 passages will be said differently, 1 word to check." In detail:
- "12/03" becomes "the twelfth of March", said the French way;
- "1,2 M€" becomes "one point two million euros";
- "3,5 %" becomes "three point five percent";
- "2025" is read as a full number;
- "M." becomes "Monsieur";
- "SDIS", a well-known French public body, is expanded to its full name, the departmental fire and rescue service;
- "CCPL" is spelled letter by letter;
- "18 h 30" becomes "eighteen thirty";
- "2e" becomes "second".
The word to check was "fils", with its two readings. The context was electrical cables, but that is for the newsroom to confirm, not for us to guess.
Two honest notes. "CCPL" is spelled out because it is not in our list of known acronyms; if it is the local association of municipalities your readers hear about every day, they would rather hear its name. And the sheet said nothing about the mayor's surname. That is on purpose: we do not claim to know how to pronounce a name our rules do not touch.
What it costs: nothing
The sheet generates no sound, calls no language model and uses no credits. It applies your pronunciation lexicon, then our French reading rules, in exactly the order the real generation applies them; anything else would show something other than what will be said. The only limit is text length, aligned with the maximum article length of your plan.
That is why it sits behind no paywall: every account can use it, trial accounts included. The lexicon itself is reserved for paid plans.
The price of a given audio is also shown before generation, on the creation screen; the estimate is covered in how many credits an audio takes.
From sheet to lexicon: fix it once
The sheet stops being a report the moment you act on it. For a homograph, a "Make it read as" button offers the possible readings; picking one writes an entry into your pronunciation lexicon. For an unknown acronym, the sheet points you to the lexicon, where you write how it should be said.
That entry applies to every article that follows, not just this one. That is the whole difference with a fix made by ear: tomorrow's article does not repeat the error.
The fix is also invisible to readers. Read-along and subtitles show "CCPL" as you wrote it; only the voice receives the pronunciation you chose.
Where it stops
- It does not replace listening. It shows what we do to the text. The voice can still stumble on a word we left alone, typically a rare surname or a place name.
- It does not guess place names. We have no phonetic engine able to work out an unusual Breton village name, and rather than display an invented pronunciation, the sheet stays silent on such words.
- It does not simulate the text preparation step. Outside the "faithful reading" mode, the text is cleaned before it is read, and that step can, for instance, expand an acronym on first use. The sheet shows your lexicon and our rules. In faithful reading mode, no model rewrites the text at all, as explained in French TTS without generative AI.
- It is French only. For a text in another language, the sheet says so instead of inventing an analysis.
Why we built it
Our position is to read one language properly rather than thirty approximately; the reasoning is on our French text-to-speech page. But a claim of quality is only worth something if you can check it. The reading sheet is the most direct way: it runs our rules on your text, with your acronyms and your names, so you do not have to take our word for it.
A simple habit is enough to get value from it: run the first paragraph of a figure-heavy or acronym-heavy article through the sheet, settle what needs settling, then generate. Every decision you make carries over to the articles that come after.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo