Hacker News Reader

Comment by bad_user | Hacker News Reader

IMHO, Google storing voice recognition samples does freak me out, especially because voice recognition is automatically activated by "OK Google" and maybe other triggers and maybe depending on version or the hardware used it's not that good - for example when my wife plays songs on her LG phone, she gets Google's voice recognition activated erroneously all the time.

So basically random bugs can get your very private conversations recorded on servers that can be under the control of state agencies and disgruntled employees, possibly stored forever? That's a reason to freak out really.

View on HN

"I will tell you my secret plans to world domination, but first need you to do something for me. OK? Google this for me and ..."

>IMHO, Google storing voice recognition samples does freak me out, especially because voice recognition is automatically activated by "OK Google"

I've had google search activate a few times by my wife watching TV in the background and once when it was sitting in front of my speakers while listening to a podcast. We've tried going back and replaying the same bits but can never get it to activate in the same place twice. Still, it is strange to be sitting there and suddenly have the Google mic pop up.

And how are we sure it isn't recording when you're NOT saying OK Google? :-)

Is there an analogue for a piece of scotch tape over the webcam for the paranoid (Zuckerbergs) of the world?

I think battery life and 4G network bandwidth are constraints for now. If it would record and stream data while in standby, battery wouldn't last for more than a few hours and would visibly consume your allocated bandwidth.

What they do is to introduce a dedicated processor that listens for these keywords and that is supposed to wake the phone up on detection. The danger of course is having that processor activate the phone on keywords other than "OK Google". And it doesn't even have to be on purpose. As I said, my wife's phone detection is totally broken for some reason.

It doesn't need to stream all the time, just record. The traffic is encrypted end-to-end anyway, you can only do an analysis on the size relative to how much time it's been idle.

This. 30* 24* 3600* 22000/(8e9) = 7.8 gigabytes per month for 24/7 surveillance at the bare minimum (AFAIK) of 22 kbps mono audio.

2.83GB.

Voice sounds fine at 8kbps. Reduce it even further if all you need is to understand what's being said.

Still, that would light up your data usage like a christmas tree.

Do you connect to wifi at least once a month?

You can also show per-app data usage over wifi on Android (it's a single click away on stock Android). If Google Now was transferring 2.8 GB of data each month, someone would have noticed. Also, the high battery usage argument.

If they were streaming spy data like that, I suspect that they wouldn't report the data usage. IMO, battery life is the strongest indicator that they aren't.

Who says it would use up your data usage? Of course this would require the network operator to allow it but it isn't anywhere near impossible to do without the user knowing.

That assumes people are speaking all the time -- the onboard processor can in fact translate most of the speech into text you know.

You only need to transmit if you detect frequencies in the range of human speech, and only in batches that can be masked with other data transfers.

I think if this were happening, people would eventually be able to tell by watching the network traffic. You may not be able to decode the network traffic, but if you run it through an analyzer like wireshark, while you say "OK, Google" and you see the bandwith increase, then if you see a similar bandwidth pattern other times you might have a clue that its recording. Of course that assumes they aren't saving to the disk and sending it later.

>and disgruntled employees,

that at least does not really worry me. Google is very paranoid over who can access customer data from the inside.

Government access is way more of an issue IMO. I would be way more creeped out by somebody going over my whole gmail history than my searches (audio or not) though.

Reply on news.ycombinator.com