What has to happen
Two separate things, and it is worth keeping them apart because one is much easier than the other.
Hearing them means something has to listen to what your computer is playing and read it back to you in your language. Nothing needs to change in Zoom.
Being heard means your translated voice has to arrive in the meeting as if it were your microphone. That is the part that needs a little plumbing: meeting apps only send audio from a device they can see as a microphone, so the translation has to be fed into one.
Hearing the other side
Install SpeakFlow Pro on your Mac or Windows PC, pick the Video call preset and press Play. It listens to whatever the computer is playing, so it does not matter whether the call is in Zoom, Meet, Teams, Zalo or a browser tab.
You get the translation spoken over the top, plus a caption panel you can drag under the video. The original is turned down automatically so the two voices are not competing — on Windows it is lowered rather than silenced, because the app hears the call through the same audio the speakers get.
This half needs no extra software at all. If all you want is to understand the meeting, you are done.
Being heard in your own voice
For the other direction the app has to hand your translated speech to Zoom as a microphone.
On a Mac, SpeakFlow Pro installs its own virtual microphone, so there is nothing to choose: in Zoom pick SpeakFlow as the microphone.
On Windows, install the free VB-CABLE, then in SpeakFlow set Your translated voice goes to → CABLE Input, and in Zoom choose CABLE Output as the microphone. There is a Play a test sentence button for checking it before the call rather than during it.
If you record thirty seconds of your own voice, the translation is spoken in a clone of it, so the other side hears something recognisably you in a language you do not speak.
Three things worth knowing before your first call
Wear headphones. On speakers, the translated voice comes out loud enough for your microphone to hear it, and then it gets translated again. Headphones end that completely. For two people sharing one laptop across a table, use the In person preset instead — it closes the microphone while the voice is playing, like a walkie-talkie.
Tell it the subject. Names, product names and jargon are where any translator goes wrong. Typing a few terms into Context, or letting the app work the topic out from the first few minutes, changes the output noticeably.
There is a lag. Usually a second or two, because a sentence has to finish before it can be translated honestly. It is worth leaving slightly longer pauses than you would normally.
What it costs
Every computer gets one free hour, no card. After that it is hour packs, counted only while a session is running, and they do not expire. A one-way video costs the same as a two-way call.
If you only ever want to watch foreign videos in a browser rather than join meetings, the free Chrome extension is the cheaper answer — it is a separate product and does not translate calls.