There could also be cues from the rest of your observable behavior even if it wasn't explicitly searched for. They have a lot of data to work with and your circumstances probably aren't unique, maybe there are signs they are aware of that you wouldn't be.

Bless you for using two different double contractions

It's not. On the one hand is however many people saying "it's obvious!" and on the other hand is no evidence of network traffic transmitting audio data. Why spend all the power to transmit audio, autotranscribe, and parse for specific keywords when they already track your browsing habits and those of your housemates?

I disagree. Consider the average internet user and how much they willingly give up about themselves online. Most of them use social media and have everyone they've ever met added on it, they post directly about what they're doing and often who they're doing it with, and they lend their engagement at things they like. They use Google for a search engine and don't block ads.

So really, for the probably 80-90%+ of the population that captures, the massive surveillance network in place just at that level is perfectly sufficient to gleam anything they might want to know. Even if someone does protect their privacy, people they're connected with still influence their profile through their lack of concern for privacy.

So really, with all that in place, what's the incentive to have a top secret voice surveillance system built on top of all that? It would destroy the market for any phone doing it if it was ever proven. Why take that risk when you can get everything you want from all those other sources instead?