I'm a firm "ad audio spying denialist", but that's mainly an "uhm, ACKTUALLY" kind of denial. Ad firms absolutely do spy, but probably not through the use of microphone audio... mainly because that's way too much information for what they actually want to do. That and every mobile OS shows very obvious warnings whenever media capture is being used, and I'm not aware of a sanctioned way to defeat those that isn't a 0day exploit that would get caught, patched, and banned off the App Store / Google Play in short order.

What I suspect is actually going on is the boring ol kind of location tracking and fingerprinting. Locations, searches, and call history. Metadata, in the language of surveillance law. Of course, the cruel joke of surveillance law[0] is that metadata is content: you can infer the content of a conversation from the circumstances of which it took place. Three-letter agencies rely on the fact that nobody understands this, and so do advertisers (of whom they pay money for data from).

[0] I'm resisting calling it the "holographic principle of surveillance" on principle.