← Blog · 11 July 2026

A few years ago, a synthetic voice gave itself away in one sentence: flat delivery, botched liaisons, intonation falling in the wrong place. You "heard the machine." Today, on a news text read aloud, most listeners can no longer tell the difference. That shift changes what a media outlet can offer its readers.

From "robotic" to neural

Old systems worked by concatenation: real recordings were cut into fragments, then glued back together word by word. The result was intelligible but mechanical — because a sentence isn't a sequence of stuck-together sounds, it's a melody.

Neural systems flipped the approach. Instead of assembling pieces, a model learns the link between text and sound signal directly, from thousands of hours of speech. It doesn't recite: it predicts how the sentence should sound.

What makes a voice "natural"

A voice's quality owes less to timbre than to prosody: rhythm, stress, rises and falls in intonation, pauses. That's what tells a question from a statement, marks a list or an aside.

A good engine reads "the president, for his part, said nothing" by marking the commas, not stumbling on them. It slows on a number, speeds up on a list, breathes between two ideas. It's this invisible work that makes you forget nobody is speaking.

Why the shift is recent

Two curves crossed. Quality first: recent models produce convincing prosody, including on proper nouns and acronyms — long the weak point of automated voices. Cost next: generating the equivalent of a read article now costs just a few cents. As long as audio was expensive and sounded fake, it stayed reserved for a few brands; cheap and indistinguishable, it can become a standard.

What it opens up for media

The audio version isn't a gadget: it's a new listening moment — car, transport, exercise — and a real accessibility lever for visually impaired or dyslexic readers. Provided production doesn't rest on manual handling, article by article.

That's the role of a service like WeDispatch: the voice is generated automatically on publish. You choose a voice (see pricing), connect your CMS once — integration takes a few minutes — and every article ships with its audio.

A building block, not a project

The real change isn't just "the voices are better." It's that audio stops being a separate editorial project and becomes a property of the article, produced effortlessly. This tool, like the others we design at Tout est Faisable, starts from a simple idea: technology should disappear behind the use.

Want to hear what this voice would do on your own articles? Book a demo.