Modes

How two-way translated conversation actually works

Conversation mode lets two people speak their own language and still understand each other. Learn the pipeline, its limits, and which apps do it best.

Published 2026-07-29 · Updated 2026-07-29

How it works

Conversation mode turn cycle
SPEAKER A TURNSPEAKER B TURNroles swap for the replySpeech in ATurn detected, language AASRTranscribe in language ATranslateA → BText + TTS in BShown and spoken to BTurn handoffAuto-detect or push-to-talkSpeech in BTurn detected, language BASRTranscribe in language BTranslateB → AText + TTS in AShown and spoken to A

Each spoken turn runs the full chain, then the roles flip and the same chain runs in the opposite language direction.

What conversation mode is

Conversation mode — also called bilingual chat or two-way voice mode — assigns each participant a language. The app listens, decides whose turn it is, translates each utterance into the other person's language, and keeps a running transcript so nothing is lost when the audio is hard to hear.

The key difference from single-shot voice translation is turn management: the app must know which language the incoming speech is in, and which direction to translate. Some apps auto-detect the language of each turn; others give each speaker a button or split the screen so each person taps their own half.

The turn pipeline

Every turn runs the same chain: turn detection, speech recognition in the source language, machine translation, then on-screen text plus text-to-speech in the target language. Errors compound down the chain — a misheard word becomes a confidently mistranslated sentence — which is why apps show the source transcript as well as the translation, so a speaker can spot a bad recognition and repeat themselves.

Latency matters more than raw accuracy here. A pause of two or three seconds between turns is tolerable; anything longer breaks the rhythm of a real conversation, so most apps stream partial results while you are still speaking.

Screen layouts

There are three common layouts. Split-screen puts each language on one half of a shared phone, rotated so the two people can face each other. Single-transcript stacks both sides in one chat-style thread. Multi-device conversation puts the transcript on each participant's own phone — that is device-to-device translation, covered in its own guide.

When to use it

Use conversation mode for genuine back-and-forth: haggling in a market, explaining symptoms to a doctor, or a long chat with a relative who speaks another language. For a single question or a menu, typed or camera translation is faster and more reliable.

In noisy places, conversation mode only works well if the app also gates its microphone — see the voice fence guide for how that gating works.

Frequently asked questions

Can more than two people use conversation mode?

Most consumer apps are designed for two. Group and classroom modes exist, but they usually rely on each participant joining from their own device.

Does conversation mode work offline?

Rarely. Most implementations need a connection for recognition, translation, and speech synthesis, though some apps fall back to offline packs with lower quality.

Do I have to pick both languages first?

Usually yes, though several apps can auto-detect which of the two selected languages each turn is in, so you do not have to tap before speaking.

Apps with this feature

← All guides