Not necessarily you or your case, but I'm still convinced that a lot of people just have confirmation bias (only noticing it when it happens and discounting the thousands of otherwise innocent ads). There's also subconscious ad effects, like you were only talking about it to begin with because your saw it somewhere because it's been spreading by weird of mouth from people who initially saw an ad

That's the gist of how it likely works; the wake word is detected by an "always on" audio DSP, but a software mode prevents the passing of microphone data back up to the SoC. I'm actually quite familiar with Amazon Echo engineering design, and they implement the "mute" feature in a manner that takes privacy seriously: the LED indicator on that button is hardwired to only turn on when the microphone is literally powered off. Thus, an Echo device can't even manage such a cheeky response, nor can a software bug or hack enable listening while the mute button is lit.

What you describe is actually how it works. If they actually sent all you say to their servers, it would be trivial to detect with a network analyser.

I will say that that's exactly how the google voice api works. Of course it's all in a black box, but that's how the documentation describes it and how it functions when making a voice app

Why listen and risk even a slap on the wrist?

Recall Target:

As Pole’s computers crawled through the data, he was able to identify about 25 products that, when analyzed together, allowed him to assign each shopper a “pregnancy prediction” score. More important, he could also estimate her due date to within a small window, so Target could send coupons timed to very specific stages of her pregnancy.

One Target employee I spoke to provided a hypothetical example. Take a fictional Target shopper named Jenny Ward, who is 23, lives in Atlanta and in March bought cocoa-butter lotion, a purse large enough to double as a diaper bag, zinc and magnesium supplements and a bright blue rug. There’s, say, an 87 percent chance that she’s pregnant and that her delivery date is sometime in late August.

These threads are always full of fucking bootlickers and paid shills.

personally I think its better to be afraid of real things that are happening than things made up by Facebook boomers.

why this particular issue fools even the most technical of people I'll never know.

"Probably" is a great way to protect your privacy haha

Back in the day when everything had to be processed server-side sure.

Now we have purpose-built hardware helping work this shit out. The devices are basically capable of handling native language resolution locally. They're no longer need to farm the data out. I still don't think they're doing this we would see it in the open source operating systems, but if they wanted to any late model cell phone would be absolutely fine parsing out your interests from your conversations. Hell, I'm sure the contents of this dictation I'm making now are being reduced and added to my social graph at Google.

I think this should be fairly easy to test yourself. Just disconnect from the WAN, say the wake word, and see if the device responds.

Someone can correct me if I'm wrong but home assistant is currently struggling with this and is processing everything on your local box because it can't do wakewords on the device.

Then how would you use it as a phone?