7 comments

  • beauzero 2010 days ago
    "If you’re in a room where a child has just fallen asleep, and someone else walks in..." considering the other person as Alexa bothers me. I appreciate the tech but the fact that Alexa/Amazon corporate is participating, even passively, in interactions with my children and myself causes me pause.
    • Invictus0 2010 days ago
      If having an always-on microphone in your home didn't already give you pause, I fail to see how this development changes things significantly.
    • PacifyFish 2010 days ago
      I think the sleeping child is the "other person" here.
    • SweetLlamaMyth 2010 days ago
      I'm not sure the takeaway here is that anybody's conferring personhood on Alexa. I think they're just pointing out that there's meaning conveyed in speech beyond just the content of the words spoken. Regardless of what words someone says, saying it in a whisper implies something about the volume that other parties should respond with. I had a coworker with laryngitis last winter, and I found myself unconsciously lowering my volume to be closer to his raspy voice. Tone and volume of speech can communicate a lot; why not allow a spoken word interface to leverage that extra information?
    • misabon 2010 days ago
      Then why buy it, dude? Do you think Alexa isn’t listening to you when you scold your kids in the kitchen? Seriously.
      • beauzero 2010 days ago
        I haven't purchased one. Good research should be encouraged. Ethical implications of implementations should be discussed openly.
        • misabon 2010 days ago
          His isn’t ethical implications discussion. This is you using every one of these feature release pages as your product review page. If nothing has changed about your position why even tell us? It’s like all those people who comment on software built with Go. “I don’t understand why Terraform uses Golang. It has no generics”

          That’s not “programming language feature discussion”. It’s obsessively off-topic behaviour.

  • new_guy 2010 days ago
    Alexa never even hears me when I shout at it, nevermind whispering!
    • jsight 2010 days ago
      That is because they haven't developed shout mode yet.
  • mattnewton 2010 days ago
    I realize I’m not the target audience of the smart speaker but I already carry a silent interface to anything I would like to do with Alexa in my pocket.

    Listening to whispers is creepy.

  • rrobukef 2010 days ago
    The article hasn't been published for a whole day and the link to the announcement (the first link) is already broken. Can anyone provide it? Thanks
    • rrobukef 2009 days ago
      The link works now.
  • krn 2010 days ago
    As a European, I cannot believe that millions of people just across the ocean are putting always-on closed-sourced internet-connected listening devices in the hearts of their homes. Maybe it's the lack of shared history? I highly recommend seeing The Lives of Others (2006)[1] to those, who want to understand why Europeans take their privacy so seriously.

    [1] https://en.wikipedia.org/wiki/The_Lives_of_Others

    • madhato 2010 days ago
      Cell phones have microphones, are closed source, connected to the Internet and carried with you into all the rooms of your home. Europeans use cell phones so why do they fear home assistants?
      • krn 2010 days ago
        Cell phones, unlike home assistants, are not designed to be recording anything by themselves. Their apps are. And the user controls both, what apps he installs on his device, and what permissions he gives to them.
        • dymk 2010 days ago
          They’re a closed source, always on device with a microphone. How do you know it’s not always listening?

          The point is, they subject you to the same threat as an Alexa. You can’t say “Europeans don’t use Alexa because it could XYZ”, and yet they use phones which could also do XYZ.

          • badrabbit 2010 days ago
            Even if they listen,you can monitor network traffic and find out for sure. The difference is a home assistant should send your recording and metadata while a smartphone with similar network activity would be suspicious.

            Anyways,this is one if those problems that should be addressed in criminal law. Unfortunately most law is retroactively created,this means society will have to face irreversible damage of technology abuse before laws are updated.

          • krn 2010 days ago
            > They’re a closed source, always on device with a microphone. How do you know it’s not always listening?

            Your iPhone might be closed-source, but Android devices can be compiled from source and fully inspected.

            • umanwizard 2010 days ago
              Ok, so what? You’ve moved the goalposts. Previously you were comparing Europe to the US, now you’re comparing hardcore privacy geeks willing to tinker with their phones (surely a tiny minority even of Europeans) to all Americans.

              The vast majority of people both in Europe and the US use stock phones from a mainstream manufacturer without changing the default OS.

              • krn 2010 days ago
                Is it that hard to see how a proprietary always-on home assistant is more invasive to privacy than a personal mobile device, which is not supposed to be listening to anything without an explicit permission?
                • umanwizard 2010 days ago
                  A home assistant is not supposed to be recording anything either, until it hears its power-up word.
            • mikeash 2010 days ago
              I’m pretty sure the cellular baseband in your phone is closed source. Coincidentally, it also has direct access to the microphone and to an internet connection you can’t monitor.
              • jschwartzi 2010 days ago
                Yeah, the baseband processor runs its own independent RTOS with full software control. Without studying, in detail, the electrical connections between the baseband and the rest of the hardware it's not possible to say that it can't record from the microphone independently of the hardware.

                It's easier to make that argument for the SIM card, which is actually not a SIM card but also a small ARM core that runs Java code. You can embed small programs into the SIM card which is something a lot of carriers do. The SIM card and baseband work in tight concert though and you should just assume that the connections of the baseband are available to the SIM card.

              • krn 2010 days ago
                Sure, but this problem is not specific to cell phones. Any IoT device which uses 2G/3G/4G is similarly vulnerable.
                • mikeash 2010 days ago
                  That just reinforces the point. Why are people so scared of an Echo spying on them but not these other things?
            • grey-area 2010 days ago
            • madhato 2010 days ago
              Android is open source but the Play Store and Play Services are closed source. If Google needed to secretly install a rootkit to listen to the microphone they could.
          • the_watcher 2010 days ago
            RE: We have an illusion of control, but do you know exactly what's going on inside the slab of black mirror? I guess this is true (for most people, it's definitely true in practice), but the difference with Echo's is that it's not just that it's _possible_ that the device is listening and recording, it's that those are proudly displayed as critical _features_ of the product. We worry about things like "the apps on my phone could be secretly recording me", but then buy Echos, Google Home's, etc, all of which require "always listening" to function at all.

            NOTE: I own an Echo and love it, despite being otherwise fairly privacy conscious (I know that these two statements are wildly inconsistent).

            • terandle 2010 days ago
              Cell phones have the same always listening features like Hey Siri, Ok Google, etc.
            • mikeash 2010 days ago
              The critical feature is that it’s always listening locally for a wake word. It doesn’t record anything until it hears that.

              There’s no real difference between “my Echo claims to only do low power wake word recognition when idle, but actually sends everything to Amazon” and “my cell phone claims not to listen to me, but actually sends everything it hears to the manufacturer.”

            • close04 2010 days ago
              One difference could be that if you discover your phone is listening and sending data that's a scandal right there while digital assistants do it by design. It simply lowers the bar to you being spied on, all else being equal.
              • judofyr 2010 days ago
                > One difference could be that if you discover your phone is listening and sending data that's a scandal right there while digital assistants do it by design.

                What do you mean? All digital assistants I know about use a wake word and they all seem to respect it very seriously:

                - Incidents where the assistant has misheard the wake word causes headlines across the Internet and there's always a reply from an official spokesperson (e.g: https://www.theverge.com/2018/5/24/17391898/amazon-alexa-pri...)

                - When Google Home Mini had a (hardware) bug where the button could cause it to record all the time, Google issued a software update which disabled the button completely. For months the button was a no-op. https://www.cnet.com/news/google-dumps-home-minis-top-touch-...

                - In the Google Home app you can see all recordings that the device has done, play them back, and see how it was interpreted. It's pretty clear that they want to be transparent about what the device does.

                All in all it seems to me that Google/Amazon is aware that a "we record everything"-scandal can be devastating for the product.

                • close04 2010 days ago
                  Maybe I wasn’t vert clear. They have the recording and sending functionality built in as main attraction. And they’re made to send that data because that’s their purpose. My phone is not supposed to do that so if I discover it does the intentions are clear.

                  And now slightly tangential, my phone will never accidentally hear the hot word, start recording, send the recording. It can’t accidentally misunderstand my pronunciation and do whatever the hotword squatter wanted to do.

                  It’s easier to abuse the one core functionality of a device than one that’s an afterthought and can be more easily disabled or stands out when it’s used. That’s because the legitimate and abusive behaviors almost perfectly overlap from a user point of view in a digital assistant.

                  Imagine being spied through an always on live streaming camera and being spied through camera that’s off most times. You may notice it powers on and sends data it shouldn’t.

              • umanwizard 2010 days ago
                Uh, no, my echo “by design” only starts recording after it hears the word “Alexa”.

                Sure it’s possible that that’s a lie, but that’s a different question.

        • wvenable 2010 days ago
          The baseband of your phone is basically an entire closed OS that runs in parallel with the OS that interact with and has complete control over all the hardware. This OS has it's own file system, programs, network access, everything.

          The Amazon echo isn't even that mysterious; it has low-power hardware to listen for the wake word and doesn't even power up the main CPU until that point. And you can always monitor the network traffic. The fears are massively overblown for home assistants. Mobile phones, by comparison, track your position in real time everywhere you are and send that information to paying 3rd parties all the time.

          • fucking_tragedy 2010 days ago
            > And you can always monitor the network traffic.

            This is meaningless if the connection is encrypted. The device can wait to send data that it's collected without telling you along with data you expect it to send to Amazon, and you wouldn't be able to tell the difference.

            • wvenable 2010 days ago
              You'd be able to tell if some traffic is being sent. If my Echo is spying on me day and in out then I'd expect a pretty constant stream of traffic from it.

              If the Echo was ever caught intentionally spying (either through analysis or someone leaking that information) that would be the end of that product line forever.

              • fucking_tragedy 2009 days ago
                In my OP, I laid out an example in which an Echo could spy on you and waits to send data along with legitimate and expected network communications.

                > If the Echo was ever caught intentionally spying (either through analysis or someone leaking that information) that would be the end of that product line forever.

                Truthfully, I don't think many people would care.

                • wvenable 2008 days ago
                  So people who do care would care and people who don't care already don't care. But trying to appeal to people who would care without any evidence is really pointless.
            • epmaybe 2010 days ago
              What if Amazon were caught doing that, though?
              • fucking_tragedy 2009 days ago
                Given how privacy scandals have played out in the past, probably nothing of much consequence relative to the size of a company like Amazon.
        • CharlesW 2010 days ago
          > Cell phones, unlike home assistants, are not designed to be recording anything by themselves.

          The assistants in phones work in exactly the same way as the assistants in home appliances, no? They start recording when they hear their voice trigger.

          Both classes of devices require trust that their maker won't invoke this functionality when it's not asked for, has hardened this functionality against bad actors, etc.

        • zamadatix 2010 days ago
          For a single app device is this distinction relevant? Or are you saying most Europeans don't install any Google, Amazon, or Facebook apps?
        • mikeash 2010 days ago
          You think they’re not. You really have no idea. You just have to trust that they aren’t constantly recording and sending everything back to the mothership. If you can trust that, why can’t you extend the same trust to a smart speaker?
          • dingaling 2010 days ago
            Distrust of a smartphone's mic can be mitigated by epoxying the microphone, or just ripping it off the board, and it's still a functional portable computer.

            But to consider further: if you're in a situation where you have to worry about your smartphone grassing on you then you probably won't keep it powered on. Or will keep it in a Faraday case or whatever. In other words you'll be considering all attack vectors and behaving appropriately.

            Whereas Amazon or Google could flip the notional 'record all' switch on their assistant devices on a whim and you'd be none the wiser.

            • umanwizard 2009 days ago
              If someone were in a situation where it made sense to put their phone in a Faraday cage, wouldn't you expect them to do the same with their Amazon Echo?
            • mikeash 2010 days ago
              If the commenter has disabled their smartphone mic as you say then I’ll give them a pass. I suspect they haven’t.

              Why can’t Google flip the “record all” switch in Android on a whim?

        • netsharc 2010 days ago
          "Controls both"? We have an illusion of control, but do you know exactly what's going on inside the slab of black mirror? There are surely exploits for both points, most probably used by your friendly government surveillance agency against whomever they consider to be their adversaries.
    • _cs2017_ 2010 days ago
      Please consider not describing your opinion as if it represents Europe.

      I'm sure there are a lot of people in Europe who disagree with you (and a lot of people in America who agree).

      Grouping people together by some vague concept of nationality or ethnicity, and then claiming that your opinion is intrinsic to that group, invites some rather unpleasant associations.

      • krn 2010 days ago
        > Grouping people together by some vague concept of nationality or ethnicity

        It's neither nationality, nor ethnicity. I explicitly raised the question of history, and how it forms different values in the society. Why doesn't the US have the equivalent of GDPR?

        • adrianmonk 2010 days ago
          The US doesn't have GDPR because of the gigantic chasm because the European and American philosophies of what the government's role in society should be.

          Europeans are more inclined to think that if there's a problem, it is very logical for the government to step in and solve it because that's government's function after all. Americans are more likely to think private citizens should solve it if possible because government should be small whenever possible.

          When it comes to privacy, an American might value privacy but think the best solution is a buyer beware approach. Such as, if you don't want to risk Amazon / Alex spying on you, don't buy an Echo. To many Americans, having to take on the entire burden of being responsible for their own privacy is the lesser of two evils because government intervention is to be avoided.

          As an analogy, consider attitudes towards medication. Some people dislike taking any medications at all if they can avoid it. If they have a headache, they will usually just deal with the pain instead of taking a pill. If their doctor wants them to go on prescription medication for high blood pressure, they will hope to find a way to manage it without drugs. Then there are other people who have a more balanced approach and have zero problem taking medications as long as they have been studied well and found safe. Americans are more like the people who avoid medications whenever possible. Sometimes they avoid things which could actually have been pretty beneficial. But they don't avoid them because they think the problem isn't a problem. It's because they don't like that type of solution.

    • Ntrails 2010 days ago
      It's not that I care about privacy particularly - it's that I can't imagine the use case for me personally. They could give them out for free and I'd not be interested in owning one.
      • blakesterz 2010 days ago
        I feel the same way. There's been a few HN threads where everyone chimed in with "here's how I use mine" and each time I read with great interest thinking "These are the people that are going to teach me something new! Something I've been missing because this thing doesn't seem all that useful to me" and I never learned everything new. I don't get the appeal at all, it's the time savings seem to be measured in seconds, and the convenience seems to be so marginal as to be not worth it for me.
      • umanwizard 2010 days ago
        The “killer app” for me is being able to say “Alexa, turn the light off” when reading a book late at night, instead of having to drag myself out of bed to flip the switch.
        • bloomer 2010 days ago
          You mean the clapper?
          • umanwizard 2010 days ago
            A particularly nice clapper that has secondary features like telling you the weather, serving as an alarm, and so on, yeah.
        • nerbert 2010 days ago
          Which Siri does anyways.
      • wvenable 2010 days ago
        I got one for free; I use it daily for setting all cooking timers and for maintaining my household shopping list. I also use it for playing music in the kitchen.

        It's pretty fantastic for all those things and I'd probably buy one now if I needed to replace it.

    • overcast 2010 days ago
      As an American, why does every topic turn into a jab from the other side of the ocean? Particularly when some of those places there are overrun by surveillance and the heavy use of cell phones. We're all in this together.
    • melling 2010 days ago
      Groundhog day: someone on HN rants about the Echo’s privacy.

      https://h4labs.wordpress.com/2017/09/27/groundhog-day-amazon...

      • bhauer 2010 days ago
        I can appreciate your fatigue with hearing small variations on the same discussion or debate in every thread about these devices, but to discount the utility of the debate is a bit narrow-minded, in my opinion.

        The value of debate, even debate that seems redundant, on a popular forum such as HN should not be trivialized. Witness the recent correction Google made with respect to Chrome and its automatic login functionality. Google did not specifically identify where they heard or read the feedback they were responding to, but any reasonable observer suspects they read a lot of it here.

        When technologists grieve over the privacy risks of in-home voice "assistant" devices, even when they (as I often do) snarkily refer to them as surveillance devices, the visibility of that debate can serve many possible ends:

        * It communicates to the vendors of these devices that technologists are suspicious of their devices. It suggests vendors need to either do better at communicating the privacy safeguards they have already created or do better at creating privacy safeguards in the first place if they want to reduce the degree of negative reaction they receive from technologists.

        * It communicates to others who do not provide these devices that there is a market for more secure devices of the same flavor, such as self-hosted in-home voice assistants that do processing locally and exclusively for the benefit of the customer. Technologists may not be saying that specifically, but it's communicated as an undercurrent.

        * It communicates to casual readers in the thread that there are risks that they should consider before purchasing and installing one of these devices. Providing clear criticism is one of the ways high-engagement people influence low-engagement people.

        * It indicates that a segment of the population will not happily stand by if R&D resources are reallocated away from traditional user interfaces to voice UIs on these cloud-connected devices. It's hard to measure, but there is no doubt that as R&D resources are allocated to voice UIs, some amount of investment is moved away from building UIs that use traditional touch, type, and click UIs. People criticizing these devices are in a small way saying "this is not the future I want." It's a way of participating in the economy and communicating demand to suppliers.

        Bottom line: Even though you and I have both "been here, done that" the value is not zero.

        • wvenable 2010 days ago
          I see where you are coming from but the debate is so stupid. With the Chrome situation, people discovered an issue, complained about it, and it was "resolved". With Home Assistants, they've never been found to spy on anyone and the manufacturers have explained in detail how they work and protect your privacy at the same time.

          The first moment that somebody finds a home assistant sending massive streams of data to the mothership then we talk about it. But until then it's just a "debate" without any facts at all just wild incorrect speculation.

          • fired_and_happy 2010 days ago
            Being reactive to a topic that's greatly discussed in this space isn't the right move either (imo). The wait until we're proven wrong is the behavior that allows entities to gain the upper hand - throughout history may I point out. I'm all for decentralized and forward-thinking companies in the AI / Automation / Assistant space. Some have been brought up here already in other comments.
            • wvenable 2010 days ago
              If you are the one claiming that home assistants are spying on you then prove it. But constantly complaining about something that is simply not happening because it might happen is ridiculous. What do you want people to do? Unplug their Echo's because they aren't spying?
    • umanwizard 2010 days ago
      What’s so hard to believe?

      Europeans care about these sorts of privacy issues far, far more than anyone else in the world, for exactly the reason you point out. It’s very abstract to everyone else.

      Maybe you’ll be proven right in the end, but for now, I love my Alexa device as it’s extremely convenient.

      • kzrdude 2010 days ago
        I think specifically Germans care about these issues far more than anyone else.
    • oulipo 2010 days ago
      This is why we are building Snips at https://snips.ai: create a 100% on-device and private-by-design Voice AI which will be open-sourced

      We did this to solve privacy issues, and empower makers!

    • verytrivial 2010 days ago
      As a European (dweller), I am also concerned by the little devices habitually carried by our political representatives. "Ho, ho, ho, Syria sounds like Siri!"[1] but seriously, just like contestants on the TV show Big Brother say, you stop noticing the surveillance after a while and start sharing things you did even realise you were sharing.

      "Surveilled By Default" is corrosive agenda to push.

      [1] https://www.newsweek.com/watch-siri-heckles-british-defense-...

  • b_tterc_p 2010 days ago
    I trust these guys know what they’re doing, but determining a whisper from a spoken word doesn’t sound like a difficult problem. I think they should have just used a logistic regression first given that it seems they’re not sure how the model works and whether the last 50 frames matter due to systemic biases in the data collection or actual trends. The graph of their posterior confuses me a bit too. Is the model predicting individually for each frame? Or is it contextualized for all previous frames up to that point? If it’s the former, their approach to picking a conclusion seems off, if it’s the latter, I would question their model.

    Edit: although I suppose the graph shown could be one of the better looking cases of certainty and not reflective of their typical result

    • abtinf 2010 days ago
      In my college phonetics class, we studied spectrograms of various human sounds. The spectrogram of whispers is very different from normal speech - it almost just looks like white noise. Forget computers, it is remarkable that even the human brain is able to process it and figure out what words are being said.
    • euyyn 2010 days ago
      It's not only a matter of telling "this was a whisper", but also of actually recognizing what in the world did you say in that whisper, no?
      • b_tterc_p 2010 days ago
        That is a separate problem from what the article describes.