So Google Records All The Microphone Audio All The Time, After All?

Posted on Oct 30, 2015 by Rick Falkvinge
Share Tweet

It seems Google does record audio from microphones all the time, despite attempts to play down the situation. The “hotword” searching – when you initiate a search by saying “Ok Google” – has been criticized before, when it was downloaded to open-source browsers running Chromium. However, major privacy concerns remain as Google doesn’t start recording when you say “Ok Google”; it was recording before you said the hotword.

Back when Google drive-by-downloaded a black box of recording code onto the Chromium open-source browser, there was a general outrage about Google code listening in to your room. The objections mainly fell into one of two categories:

  • Google is downloading a black box of code to open-source and free-software systems without permission, compromising them.
  • Google is using proprietary code to listen in to your room.

Of these, the first was considered the most grave by far, as any proprietary running code will have access to the device’s sensors – including microphone – almost by definition, and therefore it wasn’t really considered strange that proprietary code had access to black-box recording. However, it was still a privacy issue and a concern that Google had the capability to listen in to any and every room where there was a Chrome or an Android running, which – frankly – is a rather large portion of the available rooms on the planet.

Nevertheless, Google kind of roundabout apologized for having drive-by-downloaded black-box proprietary code to the open-source Chromium browser, and people attempting to shoot down the story (there are always those, and usually with the top-voted comment…) insisted that Google didn’t open the audio recording until after you had said the magic words “Ok Google”, something that is termed a hotword to begin recording, analyzing, and transcribing.

As part of Google’s transparency initiative, you can see (some of?) the data Google has stored about you. It turns out that all audio searches are saved, permanently, and you can listen to your own previous voice commands and voice searches. They’re listed in chronological order.

A screenshot from my Google Audio History. I can listen to the recordings as well as read the transcriptions. If you've been using voice search, you have a page like this, too.
A screenshot from my Google Audio History. I can listen to the recordings as well as read the transcriptions. If you’ve been using voice search, you have a page like this, too.

You’ll recall from the previous privacy discussion, that Google having access to microphones in pretty much every room gives Google the ability to eavesdrop on those each and every rooms continuously. It should be pointed out that this is a technical ability; there’s no indication this is happening, but the presence of the capability is cause for serious privacy concerns.

In that discussion, people said – no, insisted and shouted – that the fear was overblown, tinfoilhattery, and mental. “Obviously, Google only records audio after you say Ok, Google“, pundits said from all directions. But when listening to this recorded audio search off of Google’s audio search history page, something peculiar sticks out, that nobody seems to have noticed. Listen to this:

 

Did you hear it? The recording starts with me saying “Ok, Google”. That means that the above assertion – that recording starts after those words are said – is incorrect. Recording happened before those words were said. And if recording happened before those words were said, which must be the case since they’re part of the very recording, then audio recording from the microphone(s) is always active to some unknown degree. We see what Google claims is sent to Google – but of course, we have no way whatsoever of verifying this other than blind trust, and that’s in a corporation whose motto has shifted over the years from “Don’t be evil”, to “Privacy is overrated”, to “Whee, we make military robots”.

But here’s the thing. Even if Google isn’t intentionally recording at unknown cues in addition to the hotword searches – and I can’t emphasize enough that there’s no way to know this but blind trust is required – Google will still start recording audio at random times and send it to Google’s servers, when it picks up something it thinks sounds like “Ok, Google” from a conversation. This happened to me a week ago in Seoul, when I was discussing privacy concerns with EFF activists Parker Higgins and Maira Sutton (at least I believe it was that part of the conversation, but that’s beside the point here). All of a sudden, I noticed that my phone was transcribing what I was saying on its screen: Google’s audio detection had kicked in mid-conversation and it was recording the room. I was able to download that part of my conversation with the EFF activists off of Google’s Voice History page later:

 

So in summary;

We know now that Google’s audio recording does not start after you say “Ok Google”, but was active before you said it, suggesting that it’s always active to some degree.

Regardless of intentions, Google can start recording audio from a room mid-conversation anyway, as illustrated above.

This has all sorts of nasty privacy implications, and they’re not easy to work out. The way to do this properly for privacy is to make sure that all audio interpretation must happen locally, but even then, you can’t know if something is sent to a remote server if you’re searching for the wrong things as long as proprietary code is running (for instance, if you’re searching for something like methods for acquiring substances used to manufacture thermonuclear weapons, a search I dare say is never used to actually manufacture one in your basement, but which might still raise a flag somewhere).

Privacy remains your own responsibility.

About Rick Falkvinge

Rick is Head of Privacy at Private Internet Access. He is also the founder of the first Pirate Party and is a political evangelist, traveling around Europe and the world to talk and write about ideas of a sensible information policy. Additionally, he has a tech entrepreneur background and loves good whisky and fast motorcycles.

VPN Service

Comments are closed.

33 Comments

  1. a female faust

    very interesting. indeed. i would agree with the person who commented that perhaps its a streaming/caching issue,

    — except for the fact that that is giving a lot of benefits of doubt to a company caught, for instance, colluding to impersonate my operating system so as to secure my authorization, BEHIND MY BACK and via local user account password, in order to set cookies for corporate big Data web vampires like DoubleClick.

    (wrote about it here, and yes i realize Google – i mean AlphaBet, owns them now).

    which may not be, to quote a warning famously softpedaled so as not to incite unnecessary freaking out, “specifically too good.”

    3 years ago
  2. Cyberchip

    Turn it off, so that it only responds when you touch it, if you think it’s a problem. OF COURSE any device that listens for a keyword has to be listening… I thought it was obvious. Google made so many false starts when I left it on for listening, I had to turn it off. So, I did. Now it have to touch the microphone first, every time.

    3 years ago
    1. Jessica Eberl

      How do I turn it off?

      3 years ago
      1. Cyberchip

        These are generally where they’re located, depending on phone brand. If you don’t find the settings as listed below, they should be in a similar part of the device for your phone. Otherwise look up ‘yourbrand Android turn of voice recognition’ to find the settings.

        Open the Google app.
        In the top left corner of the page, touch the Menu icon.
        Tap Settings > Voice > “OK Google“ Detection.
        From here, you can choose when you want your phone to listen when you say “Ok Google.”

        3 years ago
  3. John Dearing

    I’m like this if someone can earn a paycheck by sitting or standing at a work station listening to every conversation everywhere. Which has to number in the 100’s of millions at any given time. Then good for them. I’m not sure what they can do with most of our conversation as it is. Most of it wouldn’t be admisable in a court in any case. I don’t it’s as nefarious as it is being made out in this instance.

    3 years ago
    1. Cyberchip

      Google already translates it, they’d just have to scan the text looking for words like kill president, or bomb something. lol This will probably get picked up… lol

      3 years ago
    2. a female faust

      silly netizen! they have machines for that, not humans. software. that’s how come ‘collecting’ isn’t’ ‘listening’ — *technically* —l

      3 years ago
      1. Carl Frederik

        Ha ! I known (or felt) that for a long time.

        Why does Google track what you`re searching then ?
        Also for `collecting` and `listening`.

        Damn machines. Maybe i should “go Amish”….

        3 years ago
        1. a female faust

          sorry to take so long to get back to you, especially because that is my response exactly – but with a different cutoff date. the cutoff date is really the question, or if we want one at all: the tech is not the problem, its the transparency, or rather the unilateral or asymmetric deployment of the tech.

          2 years ago
  4. Johannes Bols

    I thought OK Google was the NSA’s identifying the voice of the ISP user. When I was a contract worker for US Customs I had a security clearance. My phone was tapped more than Jim Price’s trumpt pegs during the 1972 Rolling Stones North American tour.
    When my friends and I would hear the ‘click’, meaning the line was being listened to, we launched into highly detailed gay pornographic stories. I mean REALLY detailed. Not raunch, just details, details, details. The line went ‘click’ shortly after this would start. Checkmate, fuckers…
    https://uploads.disquscdn.com/images/accc3149cd4cc2bb5a193e48d91d51c84353d58c0265c000358f333519f14c52.png

    3 years ago
  5. Stop Bush and Clinton

    Of course it listens all the time – there’s no way it can detect the “magic words” unless the microphone is on and something is analyzing what’s happening there.

    Of course we can only speculate what that binary blob does – but if it’s benign, it listens and starts acting the moment it recognizes the “magic words” – by recording (and transmitting) what it heard from the moment the recognition started returning a positive result, starting with the segment that triggered the match.

    It’s possible (and likely, simply because they can) that they transmit other stuff when they hear other interesting keywords. But this recording isn’t sufficient proof.

    3 years ago
    1. elKaiC

      Exactly my thoughts. It doesn’t need to hit a server to recognize ‘ok google’ that is being done on the device and happens when you don’t have a network connection. After the recognition occurs it might send the full recording that caused it. I’m not a privacy nut, but lately I’ve felt more and more concerned with how much access google is asking for.

      3 years ago
    2. Cyberchip

      OK google sometimes picks up stuff. I agree, as I said before I read this… How could anyone think that a black box that responds to a keyword isn’t listening for that keyword no matter how it analyses the input to see if the keyword is said. You’re right, it says nothing about what google is really doing; but, if I had no Trust relationship with them I wouldn’t be using it. It’s that simple. It’s like change the channel if you don’t like what you’re watching. I’m more worried about the stuff that might come out of a paranoid administration. As the saying goes, “Just because you’re paranoid it doesn’t mean they’re not out to get you.” It doesn’t mean they are either. Right now I just noticed a forced app in Windows 10 onenote, has a process called onenoteim (Did they say OneNote Instant Messenger?) I don’t know, but it’s always connected and listening to cloudflare in California… since they say they’re an internet protection company, what are they listening for… me to become infected… hah!

      3 years ago