Aievalu
Log inGenerate audio

AI voice generator: what it is, how it works, what it costs

An AI voice generator turns written text into a spoken voice-over in seconds, with no microphone and no studio. Aievalu is one: paste a script, pick an English or French voice, listen, download the mp3. This page explains what happens behind the button, what AI speech synthesis does well, where it falls short, and how to pay a fair price for it.

How AI speech synthesis works

Older text-to-speech engines stitched together recorded fragments of sound. You could understand them, but they sounded like a machine. A modern AI voice generator works differently. A neural model has learned, from thousands of hours of speech, how people breathe, stress a word, slow down before a comma and rise at the end of a question. It does not recite phonemes: it predicts, word by word, the melody and rhythm a real speaker would use, then renders the matching audio.

Aievalu runs on Voxtral, the speech engine from Mistral AI, a French company. The voices on offer are presets: stable, tested timbres, each in several tones (neutral, cheerful, confident and so on). You can compare them on the voices page, where every voice reads the same text so you judge like for like.

Presets or voice cloning?

There are two families of generators. Some offer ready-made voices, the presets. Others clone a voice from a few seconds of recording. Cloning is tempting, but it raises one plain question: do you have that person's written consent? Without it, imitating a voice is illegal, and most platforms will take the content down. Aievalu only offers presets for now, precisely so that every audio file you produce can be used without ambiguity. The details are in the answer about usage rights.

What an AI voice does well

Today's voices read dates, amounts, common acronyms and short dialogue correctly. They hold a steady pace across long narration, which a tired human reader does not always manage. They are available at three in the morning, and a fluffed sentence is regenerated in seconds rather than with a new take. For an explainer video, a training module or a narrated podcast, the output is usable as is.

Its limits, honestly

An AI voice does not act. It reads with a general intention, the tone you chose, but it will not cry on one specific word or burst out laughing mid-sentence. Rare proper nouns, isolated foreign words and in-house abbreviations can be mispronounced. And each generation is capped at 1,500 characters, about a minute and a half: a long text is split into passages, which is in practice the right way to work, since you fix one passage at a time. For reading documents rather than producing a voice-over, see our page on text to speech with natural voices, which covers how to prepare the text.

Steering intonation with punctuation

The main control is not a hidden setting, it is your text. A full stop is a clear pause, a comma a short one, an ellipsis a hesitation. A short sentence sounds more confident than an endless subordinate clause. Write numbers as digits, spell out the acronyms you want read letter by letter, and read the script aloud before generating: if you stumble, the voice will stumble too.

Pick the right voice for the job

The same script does not want the same voice everywhere. A few pointers:

The rules to follow

Three principles are enough. Never let a voice claim to be a real person it is not. Tell your audience the audio is synthetic whenever confusion is possible, as the EU AI Act (article 50) has required since August 2026; the mp3 files Aievalu produces carry a "synthetic voice" tag in their metadata. And respect the rules of the platform you publish on. Commercial use is included with paid credits: monetised videos, courses you sell and client work are all covered.

What it costs

Aievalu charges per character, not per minute and with no mandatory subscription. A thousand characters make roughly one minute of audio; the exact character count includes spaces and punctuation. The 100,000-character pack costs €9, valid for twelve months, which works out at about €0.09 per minute; subscriptions start at €4.99 a month for regular use. Everything is laid out on the pricing page. Before paying, the free trial lets you hear a voice on two sentences of your own text, no account needed. Voices exist in English and French only for now; the languages answer explains why.

Try with my textSee pricing