There are way too many coincidences regarding in-person conversations and Facebook ads appearing.
What people don’t understand is that voice compresses extremely well. It doesn’t need to be stereo or high-def for processing. VoIP has been around for more than two decades and worked over 128kb connection speeds for actual phone calls. With LTE and 5G, it would be almost impossible to notice compressed and encrypted voice transmissions. It would look like any other HTTPS connection because it’s just data sent for processing. It’s not a two-way call. It’s compressed audio for transcription, if it’s not already being transcribed on the device itself.