N° 048
June 3
La Phrase du Jour
A single French sentence, scored sound by sound.
Je ne comprends pas
Honest and useful — no shame in it.
Tap to record. Read the phrase out loud.
N° 048
June 3
La Phrase du Jour
A single French sentence, scored sound by sound.
Honest and useful — no shame in it.
Tap to record. Read the phrase out loud.
Notre méthode · Our methodology
01 · The model
Most speech recognition models are designed to ignore accent. Their job is to understand anyone, regardless of region, age, or first language. Ours does the opposite. We trained a deep neural network from scratch to do one thing well: hear your French the way an average Parisian would, sound by sound. It's the same family of model behind modern voice assistants and voice cloning, pointed at the opposite job. The network listens phoneme by phoneme, in context, and tells you which sounds an actual Parisian would let pass and which ones they'd flag.
02 · The training data
The network was trained exclusively on native Parisian speech — not generic French, not Wikipedia commons, not Quebec, not Marseille. Parisian. Specifically. The reason is calibration: "correct French" isn't one accent, and our users are aiming for one in particular. Training on a regional mix would teach the network to forgive features Parisians actually flag. Training on Parisian only means the score corresponds to what a Parisian's ear would do in real life.
03 · The linguistic review
Every coaching tip — the mouth-mechanics descriptions you see when you miss a sound — was reviewed by a native Parisian linguist on our team before the test went live. The criteria: would a Parisian recognize this description? Would following it actually produce the right sound? Tips that didn't pass got rewritten. The model is technical; the coaching prose is reviewed by an actual native speaker who can hear when an instruction is right.
04 · Why it's free
We put the model on the open web, no signup, because we have to be confident in it. If your score doesn't match what you'd hear in Paris tomorrow, the test is broken — and we'd hear about it within a day. The full coaching layer (daily training, mouth-mechanics walkthroughs, progress over time) lives in the Parlez app on iOS, but the scoring engine you're using right here is the same one that powers it.
Questions fréquentes · French accent FAQ