I had a hard time believing it myself and came to terms with it after reproducing it consistently over many months, despite using everything from pihole to on-device firewalls running as root.
Unfortunately, this topic tends to be routinely brushed off as coincidence and I believe it's partially because the level of surveillance underway would cause panic and outrage, if it becomes widely known. Although I have have exposure to adtech and ML, it's difficult to grasp how this sort of activity is being done at scale.
My current theory is audio may be heavily compressed prior to occasional uploads to a trusted domain, which could be obfuscated by cloudfront or similar service. It's also possible some on-device speech to text is happening, which would make the files trivial to upload. That said, on device ML itself would be a quite a feat, especially to be running on mobile devices with high degree of accuracy to isolate keywords to be used in prod, for ad targeting.
yet not even one whistleblower has stepped forward? Not one person has been able to detect something which would be so obviously detectable? Come on, man.
Unfortunately, this topic tends to be routinely brushed off as coincidence and I believe it's partially because the level of surveillance underway would cause panic and outrage, if it becomes widely known. Although I have have exposure to adtech and ML, it's difficult to grasp how this sort of activity is being done at scale.
My current theory is audio may be heavily compressed prior to occasional uploads to a trusted domain, which could be obfuscated by cloudfront or similar service. It's also possible some on-device speech to text is happening, which would make the files trivial to upload. That said, on device ML itself would be a quite a feat, especially to be running on mobile devices with high degree of accuracy to isolate keywords to be used in prod, for ad targeting.