It probably wouldn’t meet the strict legal definition, as no copy (physical or digital) is left behind for any period of time - as RAM is volatile and its contents disappear on power cycle.

There is probably some technicality in that since it doesn’t record audio, but rather just keeps a transcription - that it might not fall afoul of two-party consent?

Are you sure? We have two OLEDs (a CX from 2020 and a C1 from 2021), and both have these remotes with microphones: