Article audio on a budget: the safeguards that stop a surprise bill

Anyone responsible for a website budget asks the same question before turning on automatic narration, usually in these words: "if I set this up and forget about it, what am I paying at the end of the month?" It is the right question. Text-to-speech is billed by the amount of text read. A tool that creates audio by itself whenever you publish is a tool that spends by itself. So "how much does one article cost" matters less than "what stops the spending from running away while I'm not looking".

Here is what holds it back in WeDispatch, described from what the code actually does rather than from a brochure, followed by what no safeguard will ever do for you.

Connecting a feed creates nothing

The scenario people fear most is the archive. You add your RSS feed, it lists fifty articles, forty-nine of them years old. Should you expect fifty billed audio files within the hour?

No. The first time a feed is read, everything in it is recorded as "already seen", and that is all. Only articles that appear in the feed after that first pass trigger narration. The full mechanism is covered in connecting your RSS feed. What matters for your budget is the default: narrating a back catalogue is possible, but it is a deliberate act, never a side effect of plugging something in.

A daily cap on every feed

A feed can misbehave for reasons that have nothing to do with you. A CMS migration republishes a whole section, every article URL changes at once, or a broken feed makes old posts look new. For that day, each feed carries a cap on how many audio files it may trigger per day: ten by default. Beyond that, new items are logged but not read, and the counter resets the next day.

Ten is plenty for a blog, a company site or a local authority. A newsroom that publishes more should know the cap applies to feeds only: the WordPress plugin and the API, which send each article as it is published, are not subject to it. A site that posts three times a week will never reach the cap, except on the day something goes wrong, which is the only day it exists for.

An article already narrated is not quietly narrated again

The second hidden cost is republishing. Many CMSs send an article again every time someone saves it: a typo fixed, an image swapped, a tag added. If every save restarted synthesis, an article touched ten times would be paid for ten times.

Three rules prevent that. Text sent again unchanged is not processed again. Changed text marks the audio as "needs regenerating" without starting a new read, unless you have explicitly asked for automatic regeneration. And when you do regenerate, a targeted fix re-reads only the passage that changed, charged pro rata, instead of the whole piece.

Two switches, and the cautious one wins

Sometimes you want all automatic narration to stop: a budget to protect until the end of the quarter, a site being rebuilt, a team away for the summer. Automatic narration can be paused at two levels, on the account and on each site. The rule is deliberately blunt. If either switch is off, nothing is produced, and a site cannot turn back on what the account has turned off. The account switch belongs to whoever pays the invoice.

While paused, new articles are noted as published but not read. When you resume, the backlog is not caught up. That is a choice with a cost: articles published during the pause stay silent until you request them one by one. We prefer that to the alternative, where flipping a switch back on would trigger a bill nobody asked for. An explicit request (audio created by hand in the client area, or an API call not flagged as automatic) still goes through. The pause stops the machine, not you.

When credits run out, it stops

One credit covers an article of ordinary length, and the credit estimate is shown before anything you start from the client area is produced. The volume you buy is a ceiling, not an overage threshold. Once it is reached, the next job is refused and nothing beyond it is billed. There is no per-character overage waiting at the end of the month.

Two emails warn you first: one when 80% of the period's volume is used, another when it is gone, each sent once per period. The refusal itself gives figures (what the article needs, what is left) rather than a plan name.

Two related cases are worth knowing. A failed payment suspends production and says so, rather than letting the account nibble at what remains. And raising your volume mid-period is charged pro rata for the time left, with the extra credits added on the same basis: your renewal date does not move.

What no safeguard does for you

These barriers stop spending nobody intended. They cannot tell you whether the spending you do intend is well placed. Three decisions remain yours:

• Decide what deserves audio. Narrating everything is easy, but a three-line news brief or a legal notice page will never be listened to. If your site mixes both, connect the feed of one section rather than the whole site.
• Look at listening figures. A budget is defended with numbers, and the Listens screen shows which audio files find an audience. An article type nobody plays costs a credit for nothing every month you keep producing it.
• Plan for peaks. Elections, a big sporting event, a local crisis: the weeks you publish twice as much are the weeks your volume runs out fastest. Adjust beforehand rather than discovering the refusal on deadline night.

The short version

The real cost of an audio tool is not its price per credit, it is what happens when nobody is watching. In WeDispatch that comes down to five things: no archive read on connection, at most ten jobs per feed per day by default, no paid re-read of unchanged text, a pause that catches nothing up, and a volume that stops rather than overflows. If you run WordPress, the plugin follows exactly the same rules, because it uses the same production path and the same barriers.

Read next

• SSML for French text-to-speech: what a tag actually does, measured
• Article audio in your own player: keeping your design system and your security policy
• French text to speech for learners: a useful listening partner, not a teacher