How do A/B languages and voices work on a phone?

A wants to hear and B wants to hear specify the translation language each person wants to receive, not the language they must speak. If A wants Chinese and B wants English, A taps A's microphone and speaks; the result is translated into English for B. When B speaks, it is translated into Chinese for A. Input language is set separately in each side's settings and defaults to automatic detection.

A's voice is used for translations of A's speech, and B's voice for B's speech. Both sides may select the same voice from those available to the signed-in account. Voice previews use the current interface language, independently of either person's listening language. Both sides' usage is charged to the same signed-in account, not two separate accounts.

Tap your side's microphone to start, then tap again to stop capturing and wait for the translation. The other person taps their own microphone to reply. Starting the next turn stops any old translated speech still playing. This is turn-taking, not simultaneous capture of two speakers.