People have already done that and shown that no the device isn’t listening to you 24/7 and sending all your data out. There are plenty of papers on the subject, and it makes sense. Why record, decode and analyze all audio when your digital footprint is so much easier to compile and analyze. People aren’t random, so it’s easy to put them into statistical buckets of how to target them. Here is one reference paper (of many): https://recon.meddle.mobi/papers/panoptispy18pets.pdf