How many credits one audio costs, and how to know before you make it

Speech synthesis is billed by the character: the longer a text, the more it costs to produce. On your side, that spend is not expressed in millions of characters, which mean nothing to anyone, but in credits. Understanding what one audio consumes, and above all knowing it before you click rather than after, removes the one real nasty surprise this kind of tool can spring: discovering that an article ate two credits where you expected one.

Here is how that tally works, and where the tool shows it to you.

A credit is a block of characters

A credit covers a set number of characters, and the size of that block depends on your plan: higher-volume plans give wider credits. What does not change is the logic. The text that gets read is counted, and the moment it exceeds the size of one credit it starts a second one, even by a single character. A tiny text still costs a full credit: there is no half credit. In other words, the cost always rounds up.

That detail matters for an article sitting near the limit. Three sentences too many can tip an article from one credit to two. It is not a trap, it is the arithmetic of a usage-billed service, but it is better understood than discovered.

Two roundings, always in the same direction

There are two counts in the tool, and they are deliberately rounded in opposite directions. The cost of an audio is announced upward, as we just saw. The balance of credits you have left is shown downward. Both point the same way: never tell you something more favourable than reality. You will never be told an audio fits when it doesn't, and you will never be credited with a remaining credit you don't truly have. It is a house rule, and an automated check stops anyone from "correcting" one in the name of the other.

The estimate before you make it

When you create an audio by hand, on an article you have chosen, the screen doesn't launch anything straight away. It first reads the text and tells you three things: its size in characters, its estimated duration, and the number of credits it will cost. Production starts only once you agree. That is exactly where the nasty surprise lives, and exactly where it is removed.

The duration is given as an estimate and should be read as one: it depends on the voice you pick and the punctuation of the text, and is only truly known once the synthesis is done. The figure rests on an average reading pace recalibrated against real listens, not on a promise.

And if your credits fall short, the tool doesn't leave you facing a wall. It tells you how many characters to cut so the audio fits what you have left. A tally that merely announces "not enough credits" is useless; this one makes the trade-off possible. For the breakdown of plans and what an audio version costs line by line, see our piece on the real cost of article audio.

The automatic path asks nothing

This estimate only appears on the manual path, and that is on purpose. When narration is automatic, your site publishes an article and the tool turns it into audio on its own: nobody is at the screen to confirm anything, and asking for approval on every publication would make no sense. There, the monthly counter is the record of truth. You follow your usage in your analytics rather than audio by audio, and you set the dial once and for all by choosing your plan.

The monthly ceiling protects you, not the reverse

Finally, a spending ceiling watches in the background. It is checked before each synthesis, never after, so no bill can run away without your knowledge. This guardrail has a rule of conduct: it never cuts off a paying customer over an incident that would be ours. The one exception is the free preview offered to a visitor with no account, which does stop dead, because there is nobody behind it to answer for the spend.

In the end the credit tally hides nothing: a block of characters per credit, a rounding always on your side of caution, and an announcement before the spend whenever someone is there to read it.

See what your own articles consume

The simplest thing is to try it on your own text: the free tier shows you the estimate and the result before any commitment. Try it now, and find the full plan details on the pricing page.

Read next

• French isn't read the way it's written: what that asks of a synthetic voice
• Add the audio of your articles to a beehiiv newsletter
• French text to speech: the silences that let a long article breathe