Conversation cleary monitored and not sure how

[quarkysg]

And just think of all the conversation they listen to just to sift out keywords for advertising, every word you say is recorded, tagged, and archived.

Just curios. Is your iOS devices' Background App Refresh turned on?

Speech to text conversion should take up some processing power, so it should show up in your iOS device's Battery settings, assuming it is happening on device before uploading the converted text to the cloud. Your device's battery power would have drain quite abnormally if that's the case.

IMHO, I think that regardless of whether you spoke out loud "Portlandia" or not, the fact that it appeared in Netflix suggest that the ad algo used in Netflix targeted your home IP address and Instagram likely uses the same ad algo. The action of you speaking it out probably made it more apparent to you, but I suspect it would have happened regardless of whether your spoke it out loud.

 

[Nycturne]

Just curios. Is your iOS devices' Background App Refresh turned on?

Speech to text conversion should take up some processing power, so it should show up in your iOS device's Battery settings, assuming it is happening on device before uploading the converted text to the cloud. Your device's battery power would have drain quite abnormally if that's the case.

I don’t think this stuff does the STT on the device. It’s not horrific to send heavily compressed audio, like 32kbps mono for this sort of purpose.

 

[quarkysg]

I don’t think this stuff does the STT on the device. It’s not horrific to send heavily compressed audio, like 32kbps mono for this sort of purpose.

Would it be commercially viable to setup a STT server processing millions of uploads? Maybe it does, but it sure seems to me that they sure do go thru a lot of trouble just to sell more ads? Wouldn't this expose them to lots of legal trouble?

Besides, compressing audio also does chew up processing power, so doing so continuously would also drain battery power, and show up in the battery usage stats?

 

[Nycturne]

Would it be commercially viable to setup a STT server processing millions of uploads? Maybe it does, but it sure seems to me that they sure do go thru a lot of trouble just to sell more ads? Wouldn't this expose them to lots of legal trouble?

Besides, compressing audio also does chew up processing power, so doing so continuously would also drain battery power, and show up in the battery usage stats?

Ask Amazon, Google and Apple who all operate large scale STT services that use processing in the cloud of uploaded audio. The whole point of "Hey, Siri" is to find a clear trigger point for uploading a recording. Something like this would also be using local triggers to pick out interesting noises and only submitting that. But considering Google and Facebook are both *ad companies* first and foremost, I’d say owning an ad network and making it more valuable to get better per-impression ad revenue is quite important. I don’t think the difference of raw recording vs transcript really matters here.

Compression of audio can be done with hardware encoders these days, as you need to be able to do it quickly and at low power for things like FaceTime/Zoom/etc. Audio generally should be fairly low power, compared to the sort of things that phones are doing.

All that said, your call here that there’s a lot of data sharing going on, and the ad network itself knows what you have browsed isn’t a bad one. But at the same time, the amount of weird recordings picked up by Alexa et al and sent to these services are concerning enough that I just leave this stuff off as SOP to minimize how much gets out there.