Skip to main content

Spanish · French · German · Italian · Japanese

Learn a language by actually speaking it.

Step into a real situation — a restaurant, a market stall, a job interview — and talk your way through it with an AI that stays in character. A coach has your back while you talk, and a few minutes later you know exactly what to practice next.

Your first conversation is free: 3 minutes, no card needed.

See it in action

A partner who stays in character, in the scene you picked.

The waiter has just asked what you’d like, and the floor is yours. Your goal sits above the call, the transcript builds beside it, and the coach waits quietly in case a line doesn’t land.

A banter practice session in Spanish on the Studio plan. banter’s AI video partner, a waiter in a green apron standing in front of an espresso machine, is on a live video call. Beside her, the transcript shows her opening line: “¡Buenas! ¿Ya sabes qué vas a tomar, o te dejo un minuto más?” Above the call the learner’s goal reads “Order a starter and a main”, a status chip says “Your turn”, and a coach panel below offers help if they get stuck.
Ordering a meal in Spanish, at the Steady level, captured from a live session with the Studio plan’s AI video partner. The free conversation and the Practice plan use an animated partner.

What banter is

Speaking practice for the moment you freeze — when someone says something back.

Most apps teach you to recognize a language. banter makes you produce it, under a little pressure, in the situations you’re learning it for. Pick a scene, talk for a few minutes, and find out exactly where it fell apart — and what to say next time.

  • A conversation, not a quiz

    No flashcards, no multiple choice. You speak out loud to an AI playing a part — a waiter, a receptionist, an interviewer — and it answers what you actually said.

  • Real situations, with real friction

    The dish is sold out. The booking is wrong. The stallholder won’t budge. Scenes take turns you didn’t rehearse, because that’s the part a textbook can’t teach.

  • Feedback you can use tomorrow

    A coach beside the call when something doesn’t come across, and afterwards a score built from parts you can name, with corrections quoted in your own words.

How it works

  1. Choose your language and level

    Sign in, pick one of five languages, and set how hard you want to be pushed — from a patient partner who helps you out to one who won’t slow down. Then pick a scene.

  2. Talk your way through it

    They speak, you answer, out loud. Interrupt them, ask them to repeat, change your mind. They stay in character whatever you throw at them.

  3. Get unstuck while it matters

    If something you say doesn’t come across, your coach tells you straight away and offers a phrase that would. Stuck for words? Ask for a line.

  4. See exactly what to work on

    A score out of 100 built from parts you can name, corrections that quote your own words, and the mistakes that keep coming back.

What’s inside

The main features

8 scenes you’ll actually find yourself in

Each one has a goal to achieve and a character with their own manner — and something may go sideways once the conversation has found its feet.

  • Order a meal
  • Ask for directions
  • Haggle at the market
  • Check into a hotel
  • First-round interview
  • At the pharmacy
  • Make weekend plans
  • Café small talk

Three levels, and you choose

Set how hard they make you work before you start. It changes how fast they talk, the words they use, how much they help — and how quickly your coach steps in.

  • GentleSlower pace, everyday words — they'll help you out if you get stuck.
  • SteadyA natural pace with a wider vocabulary. You're expected to keep up.
  • BriskFast, native pace, with little slack cut for a learner.

A live coach that knows when to stay quiet

Your AI conversation partner never breaks character. Your coach sits beside the call and speaks up only when something genuinely didn’t come across — then offers a phrase that would. Stuck for words? Ask for a line.

Interrupt them. Really.

It’s a conversation, so it behaves like one: talk over them and they stop, ask them to repeat and they will, change your mind halfway through and they’ll keep up. And they wind the scene up naturally as your time runs down.

Live captions run the whole time, so you can see the words you only half-caught — and spot when you’ve been misheard.

Five languages

Every scene is written for every language — opening lines and useful phrases authored for each, not machine-translated from English.

  • EspañolSpanish
  • FrançaisFrench
  • DeutschGerman
  • ItalianoItalian
  • 日本語Japanese

Progress you can see

Every conversation is saved with its transcript and score. Your history charts the trend and rolls up the mistakes that keep coming back, so you fix the pattern rather than the sentence.

It’s your data: export all of it, or delete your account and everything with it, from your account page at any time.

A score you can do something with

Not a gold star. The number is built from parts you can name — how smoothly you spoke, how quickly you answered, your pronunciation, your grammar, your range, and whether you actually got the job done — so it tells you where the next ten points are.

  • Corrections quote your own words, with the fix and a one-line why.
  • The moments that didn’t land, and what would have worked instead.
  • Mistakes that repeat across sessions get rolled up, so you fix the pattern.
  • Pronunciation is scored by Microsoft Azure AI Speech: accuracy, fluency and completeness for each thing you say, and an accuracy score for each word. The recording isn’t kept.

Example result

Illustration
68B1

You said: “Quiero un mesa”

Better: “Quería una mesa”

Mesa is feminine — and quería is the polite way to ask.

Good questions

Do I need to be any good already?

No. The gentle level speaks slowly, uses everyday words, and helps you out when you stall — and your coach offers a phrase whenever something doesn’t come across. If you can say hello, you can start.

Will it correct me while I’m talking?

Your AI conversation partner never breaks character — that’s what makes it feel real. Your coach sits beside the call and only speaks up when something genuinely didn’t communicate. Grammar slips wait for the end, where they’re quoted back with the fix.

What happens to my voice?

Your microphone audio streams straight from your browser to Amazon Transcribe, which turns it into text, and the audio of each thing you say passes once through banter to Microsoft Azure AI Speech for scoring, which returns numbers only. banter keeps the transcript and the scores; neither banter nor Microsoft keeps the recording. You can export or delete everything we hold from your account page at any time.

How is the score worked out?

It weighs how fluently you spoke, how quickly you responded, your pronunciation, your grammar, your vocabulary range, and whether you achieved the goal in the right tone. Pronunciation is scored by Microsoft Azure AI Speech: accuracy, fluency and completeness for each thing you say, and an accuracy score for each word. The recording isn’t kept. Pronunciation makes up 15% of the Banter Score. It’s a guide to what to practice, not a certificate.

Am I talking to a real person?

No. Everyone you meet in a scene — the waiter, the receptionist, the interviewer — is an AI playing a part, and banter says so before you start. On the Studio plan it appears as a photoreal video partner; the free conversation and the Practice plan use an animated face.

How long is a conversation?

Every account gets one free conversation, and it lasts 3 minutes. On the Practice and Studio plans, each session runs up to 5 minutes.

3 minutes. One conversation, on us.

Sign in with your email — no password, no card — choose a language and a level, and pick your first scene.