Real-time voice · full duplex

Learn by talking to someone who talks back.

Not a quiz. Not a flashcard. A face and a voice on the other side of the table — and it hears you the moment you speak. It answers, it interrupts, it lets you interrupt it.

When you get something wrong, it doesn't grade you afterwards. It says: “that works — a native would say it like this.”

Start a conversation How it teaches

Full duplexspeak over it, it speaks over you
Live facespick who you're talking to
Any languageor your own, sharpened

Teacher

listening voice: warm · mid-range

Tap a face below to change who you're talking to.

Pick your teacher.

Every face is a different person: different pace, different patience, different accent. Same method underneath.

The correction happens inside the conversation.

That is the whole idea, and it is the part every other app gets wrong. You don't finish a chat and then get a report. You say something, and the person in front of you reacts to it — the way a real teacher listening across a table reacts.

It meets you where you are

First minutes are a conversation, not a test. It finds out what you already understand and starts one step above it — enough to follow, not enough to coast.

You talk more than it does

It asks, you answer, it pushes. When you can't find a word, it supplies it and moves on, so the sentence never stalls. Stalling is the enemy, not mistakes.

It repeats it back, fixed

You say it slightly wrong, it answers using your sentence the right way. You hear the correction without being told you were wrong. That is how people actually absorb a language.

It knows what to bring back

The words and mistakes from last time are still on the table next time. Nothing you struggled with gets silently dropped.

You

“Yesterday I go to the market and I buy a… how do you say…”

Teacher

“A scale? Sure — yesterday I went to the market and bought a scale. Was it for the kitchen?”

You got the past tense wrong and stalled on a noun. You were corrected once, quietly, and given the word — then the conversation kept moving. Nobody said “incorrect”.

Two ways to use it.

You don't speak it yet

A0 → B1

Slow, graded speech — it comes down to your level instead of you climbing to it.
Grammar arrives attached to the sentence you just tried to say, never as a rule list.
Pronunciation is fixed on the words that actually blocked you.

Goal: hold a five-minute conversation without switching to English.

You already speak it

B2 → native

Better wording, not correction: “fine — but here's what a native would actually say.”
Register: the difference between writing to a landlord and texting a friend.
Filler words, hedging, rambling, over-explaining — the habits that make people sound unsure.

Goal: sound like the person you are, in the language you're speaking.

The voice belongs to the face.

Pick a teacher, then pick how they sound. One voice at a time, matched to the person you're looking at — no separate menu to keep in sync.

More voices unlock per teacher once a conversation is open.

Sit down and start talking.

Headphones in. One click. It greets you first.

Open a conversation

Runs in the browser. Microphone permission is the only thing it asks for.