My best guess is that I know one of them uses Facebook. Apple phones. Facebook, Uber, and a few others have had pretty deep access to APIs not accessible to other software companies. Sometimes they're caught like when Uber was caught using a screen scraping API. Sometimes they aren't. The other guess that glues it together is that Facebook has indeed scraped audio to text for a long time. It was almost 10 years ago that I had the EE conversation.
Google and Meta pay Apple money to gain access to their user metrics. It's likely symbiotic relationships. Facebook once had hooks directly in iOS. Likewise, the little mic/video indicators the OS displays when they are "active" are completely software-controlled and can be overridden.
At a time, I worked at a company that had(has) deep access to other aspects of iOS. Apple always required the source code is available to them so they could inspect it. I doubt that has changed. It also means they would be complicit. External tools wouldn't really be able to figure this out. For someone to black-box this they'd need a jailbroken iPhone and some specialized tooling or MITM decryption capabilities.
Not to sound hyperbolic, I'm connecting dots with no evidence, it's pure speculation. The compute seems to be there and with no regulation in source code, anything goes, if you want money bad enough. Especially with the mad dash every tech company has been on for the last 20ish years to harvest everything they can, ever since smartphones became powerful and commonplace enough.
And save what money you can now, you can do a recast/refi and drop the principle with that saved money when you're doing it.