French quality: what it actually means for a voice tool
Our stance, stated plainly: why we would rather read one language well than thirty badly. And the list, with no spin, of what we do not do, with the reason behind each no.
"Quality" is the emptiest word on a spec sheet. Everyone claims it, no one proves it, and a seasoned reader has learned to stop seeing it. So we would rather say what it means for us, concretely, and prove it the only honest way there is: through the list of what we refuse to do. A tool is judged as much by its noes as by its yeses, because a deliberate no costs the one who makes it, while a yes to everything costs nothing and means nothing.
Our stance fits in one sentence: we would rather read one language well than thirty badly. Here is what that choice implies, and what it forbids us.
What we do not do, and why
We do not read thirty languages. Five only: French, English, Spanish, German and Italian. Why this no: reading a language well takes preparation work specific to its grammar, its numbers, its acronyms, its liaisons. That work does not replicate for free from one language to the next. A catalogue of thirty languages is reassuring on a spec sheet and says nothing about the quality of any of them. If you publish in Portuguese or Arabic, we are not the right tool, and we would rather tell you that than sell you one more flag.
We do not offer a voice that reads everything. Each of our 163 voices reads a single language, the one it sounds native in. Why this no: a versatile voice that switches between languages always betrays its origin, and a French listener hears the accent within three words. We left those voices out, even when they sounded good, because naturalness beats versatility.
We do not claim to be sovereign end to end. Text preparation happens in the European Union; voice synthesis, for now, does not. Why this no: as long as the voice chain is not European from one end to the other, the so-called sovereign offering stays greyed out, and we do not tick a box we cannot honour. A provider who answers "everything is in Europe" without separating the two steps deserves a harder question.
We do not keep every beautiful voice. We settled on a single reading engine, chosen because it timestamps every word in the audio file. Why this no: that word-level alignment is what makes possible the text that highlights as it is read, search inside the audio, subtitles and targeted correction of a passage without redoing the whole thing. Some better-sounding voices cannot do that alignment: we switched them off rather than promise a highlight they could not deliver. We chose the feature over half a tone in the voice.
We do not bill per character. The industry quotes prices per million characters, which helps no one decide, because no one listens to a million characters: you listen to an article. Why this no: we reason per article and per monthly volume, so the price relates to what you actually produce. What "free" really covers, from one provider to the next, we detailed in what a free French text to speech actually costs.
We never count mass as an argument. You will not see us write "thousands of articles narrated" or "the number one solution". Why this no: the count says nothing about how well a single article is read, and it is a single article that your reader hears. Mass impresses; it proves nothing.
Why this stance, and not another
One could object that it would sell better to tick every box. True, and that is precisely why we do not. A French-language publisher does not judge a voice tool on the number of languages or voices on display, they judge it on how their own article is read, with its town names, its amounts, its acronyms. That is where trust is won or lost, and it is the only ground on which we accept being compared.
French quality, for us, is therefore not a label or a promise. It is a series of refusals: fewer languages, fewer voices kept, no box ticked in error. What remains after those refusals is a tool that does a small number of things well, and says so.
The foundation of all this, the preparation of French reading, is detailed on the French text to speech page. And if you want to test us on our own noes, the checklist of six questions to ask before choosing applies to us as much as to any competitor: ask them, and be wary of anyone who answers yes to all six.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo