Put the computer between you. Each of you speaks normally; the app works out which language it just heard and says it in the other one, out loud.
No headsets, no passing a phone back and forth, no taking turns pressing a button.
The app detects the language of each sentence and translates it the other way, so neither of you has to switch anything.
An echo guard closes the microphone while the translation is playing, like a walkie-talkie, so it never translates itself.
A licence keeps working without a connection for a week, which matters when you are travelling.
This turns on the echo guard and the two-way detection together.
What you speak, and what they speak.
Leave a beat at the end of each sentence so it knows you have finished.