Reading units of measurement aloud in French: the table of traps
km/h, m², °C, kWh, 3,5 cl, 250 Go: a unit does not read the way it is written, and the symbol carries neither its gender, its plural, nor its word order. The table of units that trip up French speech synthesis, and what handles them correctly.
A unit of measurement is an abbreviation that looks like a word. In French, « 90 km/h » does not read as "ninety k m slash h" but as « quatre-vingt-dix kilomètres heure ». « 25 m² » is not "twenty-five m two", it is « vingt-cinq mètres carrés ». « 4 °C » is not "four degree C", it is « quatre degrés Celsius ». Each of these forms hides three decisions the symbol never states: how to expand the abbreviation, how to agree the plural, and in what order to say the parts when there are several. This is where a generic voice stumbles often, and where the mistake is heard at once, because a badly read measurement is one you cannot write down by ear.
The table of units that trip up the machine
| Written form | Expected reading (French) | Common error |
|---|---|---|
| 90 km/h | quatre-vingt-dix kilomètres heure | "ninety k m slash h" |
| 25 m² | vingt-cinq mètres carrés | "twenty-five m two" |
| 4 °C | quatre degrés Celsius | "four degree C" |
| 3,5 cl | trois centilitres cinq | "three point five c l" |
| 250 Go | deux cent cinquante gigaoctets | "two hundred fifty G o" |
| 500 kWh | cinq cents kilowattheures | "five hundred k W h" |
| 1,80 m | un mètre quatre-vingts | "one point eighty m" |
| 12 mm | douze millimètres | "twelve m m" |
| 60 min | soixante minutes | "sixty min" truncated |
| 100 g | cent grammes | "cent g" or "cent gramme" without agreement |
The right-hand column is not a guess. These are the readings you actually hear from voices that predict sound without an explicit model of French conventions. Each row is a test you can put to any tool, ours included, in two minutes: read the left column aloud through it, listen, compare with the middle column.
Why a unit resists a naive reading
The first trap is expansion. The symbol "km" is not a word, it is an international convention that condenses "kilometre". An engine that treats letters as letters will say "k m", spelling them out. It has to recognise the string as a unit, look up its full name, and say it. This mechanism is a direct cousin of the one described in reading numbers aloud in French: a bare number and a number followed by a unit do not follow the same reading rule.
The second trap is plural and gender. « 100 g » reads « cent grammes », with an "s" that never appears in the symbol. « 4 °C » gives « quatre degrés Celsius », masculine plural. The agreement follows the number, not the symbol, and a tool that copies the symbol as written produces « cent gramme » or « quatre degré », forms that sound instantly wrong to a French ear. The same agreement logic applies to the everyday abbreviations covered in the article on abbreviations in speech synthesis.
The third trap is order and composition. « km/h » is not "kilometre divided by hour", it is « kilomètres heure », and the slash is not pronounced. « m² » does not read "m two" but « mètres carrés », the superscript becoming an adjective. Finally, the decimal shifts position in speech depending on the unit: « 1,80 m » is commonly said « un mètre quatre-vingts », the decimal part moving after the unit, exactly like the cents after the euro in prices and currencies.
What happens before any sound comes out
None of these cases is solved by a nicer voice. They are solved by a normalisation step that recognises a string as a unit, looks up its name, agrees the plural with the number, expands the prefixes (kilo, giga, centi) and puts compound parts back into French word order before synthesis. It is rule work, not timbre work, and it is the whole logic of the French text to speech page: the essential decisions are made upstream. A beautiful voice fed badly will read "twenty-five m two"; an ordinary voice prepared well will read « vingt-cinq mètres carrés ».
What WeDispatch does with it, without overpromising
The common units in the table, distances and areas (m, km, m², km²), masses (g, kg), volumes (l, cl, ml), temperatures (°C), durations (min, h) and the usual data units (Ko, Mo, Go), are expanded and agreed by the normalisation applied to every article, in both reading modes. The plural agreement follows the number, prefixes are expanded, and a compound unit like km/h reads in French word order.
Some cases remain where the choice is not universal. A rare unit, an abbreviation specific to a sector (a technical unit, an internal notation), or a unit whose reading depends on context: guessing wrong would be worse than staying cautious. For a unit that recurs in your content and must always read the same way, the pronunciation lexicon lets you fix the expected form once and for all, with nothing to regenerate. We prefer to put it that way rather than claim a machine always guesses right: it applies solid rules to the frequent cases and leaves you in charge of the rest.
The test to run before you choose
Take a real page from your site where measurements matter: a spec sheet, a recipe, a scientific article, a set of instructions. Find a compound unit (km/h, m²), a prefixed unit (kWh, Go) and a decimal measurement (1,80 m, 3,5 cl). Have the tool you are evaluating read those passages, and listen only to those points: could you write the measurement down by ear without asking for a repeat? That is the only criterion that counts, and it predicts far more than a general listen.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo