There are three ways to get a Zoom or Teams call into a transcript, and they fail in different places. Knowing which one a tool uses tells you more about how it will behave than any feature list.
The service dials into your call as a participant, and the recording happens on their servers. This is what most of the well-known note-takers do.
It works well, and it has two costs. Everyone in the meeting sees a participant they did not invite, which changes how some conversations go. And a large number of workplaces block these outright — if you have ever watched a note-taker sit forever in a waiting room, that is what happened.
The simplest thing an app can do, and the easiest to get wrong quietly. Your voice is recorded properly; the other people are recorded as sound coming out of your speakers, after your meeting software's echo cancellation has spent the entire call trying to suppress exactly that. It sounds acceptable in a quick test where you are doing the talking, and it falls apart in a real meeting where you are mostly listening.
macOS can hand an application the audio that other applications are playing. A recorder using that route gets the far side of the call at full digital quality, direct, with no room, no echo, and nothing suppressed. Nobody joins the meeting; the machine you are already using does the work.
This is the approach you want for calls, and there is one further decision inside it that matters more than people expect.
A recorder taking both your microphone and the system audio can either mix them into one file or keep them as two tracks.
Mixed is simpler and loses something you cannot get back: once the two voices are in one waveform, working out who said what becomes a machine-learning problem. Speaker diarization is decent now, and it still stumbles on crosstalk, similar-sounding voices, and anyone with a poor connection.
Kept separate, the same question is answered by the file layout. Track one is you. Track two is everyone else. In a two-person call that is a complete answer, permanently, with no model involved. For a group call, diarization only has to sort out the far side — a much smaller job than starting from a mixdown.
Four questions, and most product pages answer none of them directly.
I build a local meeting recorder that does the two-track thing, for people who take their notes by hand on a reMarkable tablet — the finished note comes back to the tablet with the handwriting placed beside what was being said. There is no public installer yet. What it is and where it honestly stands.