Paste text to turn it into audio, when the source has no address

The usual way to produce the audio of an article is to give its address: you paste the URL, the tool fetches the text on the page and turns it into a voice. But not everything you want to narrate has a public address. An internal report, a press release still under embargo, a newsletter going out by email, a text you are still writing: none of these has a URL to give. For those cases, there is a second route, as direct as the first: you paste the text itself.

Two entries to the same audio

On the generation screen, next to the field where you paste an address, a link leads to a text box. You drop your text in, plain, with no particular formatting. From there, everything happens exactly as for an article pulled from a URL: you confirm, you choose the voice, and the audio is built. The pasted text joins the very same second step as the address, which is not a detail of ergonomics but a deliberate choice: two different confirmation screens would sooner or later drift apart, and one of them would end up less well maintained than the other. One path, one thing to understand.

This second route opens uses that a URL closes off. A newsletter is often prepared in an emailing tool, not on a web page: you paste it and you have the audio version before it even goes out. A working document, an internal note, a speech, a chapter in progress: any text in front of you becomes narratable, whether it is published or not.

The estimate is worked out in front of you, asking nothing

There is a small thing that changes the whole feel of it: the character count updates as you type. You paste a text, you see its length at once, and if it is too long, you cut before sending anything. That estimate costs no round trip to the server, and that is only logical: the text is already there, in front of you. Counting it is a calculation, not a question to put to a distant machine. Asking the server for the length of a text you have just pasted would be a call for nothing.

It is the same spirit as the one described on the French text to speech page: put the voice to work only on what truly has to be read, and settle everything else as close to the user as possible. Seeing the length before you launch the generation is also the best way to anticipate the cost of an audio version, which depends directly on the number of characters.

A button that promises only what it will keep

There is a minimum number of characters below which a text is not narrated: too short, it does not produce a useful audio. That minimum is not a value invented for the screen: it is exactly the one the server applies when it processes a URL. As long as your pasted text has not reached the threshold, the generation button stays inactive, and it shows you so. It is a simple rule we hold everywhere: an active button that will later be refused is a lie by invitation. Better a greyed-out button that tells the truth straight away than a clickable one that disappoints after you send.

The title of the audio, too, is derived from the text: the first sentence serves as a marker, bounded so it does not overflow. It is not a summary, it is a label to find the audio again in your list. The same rule holds for a pasted text and for an imported document, because it is the same module that carries it: two copies would have ended up producing two different titles for the same text.

Who it really serves

Pasting is not a fallback feature, it is a full entry point in its own right. It serves anyone who produces text before it goes online, or text that will never go online at all. A newsroom that wants to hear an article before publishing it, to judge its rhythm. A local authority narrating a restricted report. An author testing how a chapter sounds. In all these cases the public address does not yet exist, or never will, and that is no reason to go without the audio version. If your needs run toward automating large volumes, the API is where to look instead; for one text at a time, right in front of you, pasting is the shortest move there is.

Read next

• The diaeresis out loud: how French text to speech handles Noël, maïs and aiguë
• iPhone, Ikea, L'Oréal: how a French voice reads brand names
• Reading times and opening hours aloud in French