Talk

Talking across a language barrier. You record a message in your own language — the other person hears it in theirs, spoken aloud, and reads both texts. For a conversation on the spot, one phone between two people is enough. No shared language is needed at any step.

Voice messages

Hold the button and speak, or type when a room is too loud or a street name is easier spelled than said. The message arrives translated, in the listener's own language.

One message, several people

Send one recording to up to 20 people at once — each of them hears it in their own language.

In your own voice

With a voice clone set up, the translation is spoken in your voice — a clone built from Portuguese speech can read Russian in it. Optional, and always labeled.

Live, face to face

One phone on the table between two people. The far half of the screen is upside down, so the other person reads it the right way up from across the table.

How live mode works

Open Talk and start the live conversation: pick the two languages and, if you wish, the voice that speaks for your guest.
Lay the phone between you. Each half has its own talk button — whoever speaks holds theirs, and the phone reads the translation aloud and shows it on the other half.
The session takes the whole screen and keeps it awake; navigation and notifications step aside. To finish, tap twice on the edge by your own half — a deliberate gesture a stray swipe cannot trigger.

Your guest needs nothing

No account, no app, and no consent to anything — because nothing your guest says is kept. No transcript, translation or audio survives the conversation; only the minutes count toward your quota. And if they talk into the wrong half of the phone, the direction corrects itself instead of wasting the turn.

Your voice, on your terms

The cloned voice is strictly optional and always yours to remove:

It begins with an explicit consent step — never a pre-ticked box. A voiceprint is sensitive personal data, so your consent is recorded, dated and provable.
The clone builds itself from your own spoken messages: about a minute of speech makes the first version, two minutes refine it — and then the recordings are deleted. A progress bar shows how close it is.
Only your own speech is ever used: typed messages and live conversations are never clone material.
“Delete my voice” removes the clone and its samples and withdraws the consent.

Or pick a voice instead

A voice picker chooses how your messages sound: your clone, one of the curated voices, or the automatic default — a voice native to each language, so Portuguese sounds Portuguese. An explicit pick always wins over the clone.

Always labeled

Every synthesized message carries a visible badge — “AI voice”, or the sender's cloned voice when one spoke it. A synthetic voice never passes itself off as a recording.

What is kept — and what is not

The two texts — what was said and what was delivered — remain as the record of the conversation. The audio itself expires after 30 days; the message still reads.

Translation can be switched off per conversation (“as said”). A message that goes out untranslated is clearly marked for both sides, so nobody mistakes it for a translation.

Getting started

Open Talk in the menu, or use the Voice note action on someone's profile.
Hold the talk button and speak — or switch to the keyboard and type.
To reach several people at once, choose “Say it to several people” and pick the recipients.

30 minutes of spoken audio per month are free — voice messages and live conversations combined.