Punctuation and breath: what a comma actually does to spoken audio
A comma, a semicolon and a period do not produce the same pause. The same paragraph punctuated two ways, the measured difference in length, and why badly punctuated text gives itself away out loud.
On the page, punctuation is something you see. Out loud, it is something you hear. A synthetic voice does not guess where to take a breath: it reads the score you hand it, and that score is made of your commas, your periods and your semicolons. Text that is loosely punctuated does not become wrong on screen, because the eye jumps back and forth and repairs it for free. But handed to a machine that reads line by line and never re-reads, the same text starts to run together, or stops in the wrong place. That is the kind of flaw you only catch by listening.
Punctuation is the score of the silences
When a synthesis engine meets a comma, it does not add a sound: it adds a silence, a short one. A period, a longer silence. A semicolon, something in between. A colon, a suspension that announces what follows. The speaking rate of the words does not change from one mark to the next; what changes is the length of the gaps between groups of words. Those gaps are not empty. They are what cuts the sentence into units of meaning, exactly like a human reader's breathing. Take them away and the listener hears a stream of words with no joints, which they then have to re-segment in real time. It is tiring, and after two minutes they drop off.
The same paragraph, two ways of punctuating it
Take one text and change only its punctuation, not its words.
Careful version: "The town council, meeting on Tuesday night, passed the budget. Three members abstained; two voted against. The mayor, visibly relieved, closed the session at eleven."
Flattened version, same words run together: "The town council meeting on Tuesday night passed the budget three members abstained two voted against the mayor visibly relieved closed the session at eleven."
Count the silences. The first version places four commas, one semicolon and two internal periods. With the pause lengths an engine typically inserts (on the order of 0.25 second for a comma, 0.45 for a semicolon, 0.65 for a period), that adds up to roughly 2.7 seconds of silence spread through the passage. The second version offers no internal footholds: the voice reads it in one breathless run, and those 2.7 seconds vanish. Same words, same voice, same rate: the careful version runs close to three seconds longer, and those three seconds are precisely what makes it understandable. Run the test on your own text and the exact figure will depend on the engine, but the direction never changes: silence is content.
Not all pauses are equal
The opposite mistake exists too: too many commas, or commas in the wrong places, chop the reading into a panting, out-of-breath delivery. A comma separates, it does not conclude; using it where a period belongs produces a sentence that keeps gasping for air. A semicolon marks a strong link between two independent clauses: to the ear it is a pause that says "more of the same is coming", and an engine that treats it as a plain comma loses that signal. The colon is the most interesting mark: it does not stop, it announces, and a good reading renders it with a slight upward suspension, the way you lift your voice before a list. When you wonder how a tool handles these marks, that is exactly the question to put to it on your own text, and it is the heart of what the French text to speech page prepares: punctuation is not decoration, it is a reading instruction.
What this changes for whoever writes
The good news is that none of this asks you to learn a technical language. You simply punctuate for the ear as much as for the eye. A sentence you could not read aloud without running out of breath is a sentence to split. A list is better announced by a colon than tossed in after a comma. A semicolon, when it is earned, is audible and gives structure. These are the same rules that make any text pleasant to read, except that in audio they stop being optional: the listener cannot glance back to recover a badly cut sentence.
This is also why reading quality is not only about the timbre of the voice. Two other places deserve the same care: numbers read aloud, each of which has an expected spoken form, and the way a tool handles the hard cases you can check for yourself, following the method in how to evaluate a neural voice. If you want to judge a tool seriously, have it read one well-punctuated paragraph and the same one badly punctuated: the gap you hear will tell you more than any polished demo. What matters plays out on your real text, commas included.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo