AI voices and accessibility: make your content audible

12 October 2026 · 7 min read · Figures checked on 12 October 2026

A text you can listen to reaches readers the written page leaves behind: people with dyslexia, low vision, fatigue, no time, or simply on the move. Synthetic voices make that audio version affordable for any content, from a user manual to a newsletter. On one condition: do not mistake it for a complete accessibility solution. Here is what it brings, what the law expects, and how to produce it properly with the English and French voices of Aievalu.

Who it actually helps

People with dyslexia first: reading a long text takes an effort that listening removes. People with low vision, who often use a screen reader but appreciate a well-produced audio version for long content. People whose cognitive load is already saturated, at the end of the day or under stress, for whom a calm voice lands better than a dense page. And everyone, walking, cooking, driving. An audio version is not reserved for one audience: it is one more door into the same content.

What the law expects

In France, the RGAA, the general accessibility framework, applies to public bodies, public-service delegates and companies with a turnover above 250 million euros. It covers the accessibility of the sites and apps themselves: structure, contrast, keyboard navigation, alternative text for images, screen-reader compatibility.

At European level, Directive 2019/882, the European Accessibility Act, has applied since 28 June 2025 to many private-sector digital products and services: e-commerce, banking, e-books, electronic communications, transport. It sets accessibility requirements for new products and services, with a transition period until 2030 for some services already on the market.

In both cases the key point is the same: an audio version is a complement, not a substitute. It does not replace structured text, alternative text for images or accessible navigation. It adds to content that is already accessible and makes it easier to consume. For a small publisher outside the RGAA's scope, it is often the most visible and least expensive gesture.

Practical uses

The audio version of an article or newsletter, placed at the top of the page with a simple player. Product instructions, read while you handle the object. The onboarding pages of a service, for new users who prefer to listen while they discover the interface. A local authority's practical guides, a venue's safety instructions, house rules. The common thread: a written, stable text that many people must understand. For long teaching content, the e-learning voice-over guide goes further; for slide decks, voice-over for presentations.

Producing an audio version that helps

A bad audio version tires as much as a bad text. A few rules change everything. Read the headings aloud: announce each section, the listener has no table of contents in front of them. Write numbers out in words when the reading matters, especially amounts and dates. Cut paragraphs into short sentences, one idea each. Produce one file per section rather than a twenty-minute block: people can skip, go back, resume. Choose a neutral, steady voice, Paul Neutral or Jane in English, Marie Neutre in French, rather than a cheerful tone that wears thin over time. The full writing rules are in the article writing a script for an AI voice, and how the engine works is on the AI voice generator page.

Put the audio player in the same place on every page, with an explicit button, "Listen to this article", and state the duration. Say that the voice is synthetic: it is honest, and the European AI Act requires it when content could be mistaken for the real thing. Every Aievalu MP3 also carries a technical marker in its metadata.

What an AI voice does not replace

A screen reader reads the whole interface, buttons, menus, forms, and follows the user as they navigate; an audio version only reads a fixed piece of content. Captions remain necessary for video. Alternative text, contrast and keyboard navigation belong to how the site is built, not to the audio. And a synthetic voice, however natural, does not fix a confusing text. Do the text first, then the audio. For reading long texts in general, see text to speech with natural voices.

Where to start

Pick a page that many people open and few read to the end: a FAQ, a getting-started guide, a practical sheet. Read the text aloud and shorten whatever leaves you out of breath. Generate one section, listen to it in full, fix the numbers and acronyms that were misread, then chain the other sections in the same voice. Publish with a visible listen button and ask two or three readers concerned, dyslexic or visually impaired, for feedback: their opinion beats any rule. Then measure listens for a month before extending the approach to the rest of the site.

What it costs

A 1,200-word article is about 7,000 characters, €0.63 with the €9 pack of 100,000 characters, valid for twelve months. A weekly newsletter narrated all year fits in that same pack. Credits are only charged when the audio is produced, following the counting rule, and submitted texts are not stored, only a technical hash is. The plans are on the pricing page; the free trial lets you hear two sentences of your own text, no account needed, before you decide.

Sources

This article is informational, not legal advice. To know whether your organisation falls under the RGAA or the European Accessibility Act, consult the texts cited or a specialist adviser.

Try with my text See pricing