3 ms·

In theory, the device could only listen for its trigger word using local processing, which then and only then would trigger a cloud upload to process the follow

In theory, the device could only listen for its trigger word using local processing, which then and only then would trigger a cloud upload to process the following phrase. The user has implicitly indicated "I'd like a cloud lookup on this please, and I understand what I say next is going somewhere out of my control." If you go with continuous recognition and upload, you then have everything you say being removed from your control. You'll see situations where cloud providers' records are subpoena'ed like that murder case with Alexa. [1] You'll also see cloud providers say "if you want to hav a private conversation, you best leave the room." [2] That's terribad for everyone. 1. https://www.engadget.com/2017/03/07/amazon-hands-over-alexa-data-after-murder-suspect-gives-the-okay/ https://www.engadget.com/2017/03/07/amazon-hands-over-alexa-... 2. https://www.cnet.com/news/samsungs-warning-our-smart-tvs-record-your-living-room-chatter/ https://www.cnet.com/news/samsungs-warning-our-smart-tvs-rec...

bootlooped 9y ago
I don't think they would move to continuous recognition and upload. I think it would simply be listening for a different keyword, which it would recognize on-device, and then upload the audio following that. It would be pretty insane if every android smartphone was continuous uploading audio to Google's servers for processing. Can you imagine the processing requirements, the bandwidth load, the battery drain?
- imglorp 9y ago
  Mobile, yeah, but going by the above two references, both the Alexa and Samsung examples _are_ doing something with continuous upload or they wouldn't be in the news.
  - EpicEng 9y ago
    If you had bothered to read the article... >While Alexa is constantly listening, it only records data after the "Alexa" or "Amazon" commands are given. However, the Echo speaker was reportedly streaming music near the tub, and investigators believe it could have been awakened intentionally or accidentally. If so,the presence of a recording could tell them if Bates was awake when he said he was sleeping, for instance. It only sends data after hearing its wake up words, just like everything else. If it were constantly streaming they would be caught as soon as someone decided to fire up wireshark (i.e. very soon after release.)
    - joshumax 9y ago
      I even ripped apart an Echo dot a little while ago, de-soldered the NAND from the device and dumped the FireOS-based ROM to take a peek myself. It is indeed performing audio processing for the wake-up word entirely locally.
      - imglorp 9y ago
        This is the most important comment here. Some independent verification. Worthy of a blog post?
monkeynotes 9y ago
Well, it looks to me that all we have at the moment to get upset about is some speculation. We need to make a bunch of assumptions to get to 'Google is logging everything we say'.

bootlooped 9y ago

I don't think they would move to continuous recognition and upload. I think it would simply be listening for a different keyword, which it would recognize on-device, and then upload the audio following that. It would be pretty insane if every android smartphone was continuous uploading audio to Google's servers for processing. Can you imagine the processing requirements, the bandwidth load, the battery drain?

imglorp 9y ago
Mobile, yeah, but going by the above two references, both the Alexa and Samsung examples _are_ doing something with continuous upload or they wouldn't be in the news.
- EpicEng 9y ago
  If you had bothered to read the article... >While Alexa is constantly listening, it only records data after the "Alexa" or "Amazon" commands are given. However, the Echo speaker was reportedly streaming music near the tub, and investigators believe it could have been awakened intentionally or accidentally. If so,the presence of a recording could tell them if Bates was awake when he said he was sleeping, for instance. It only sends data after hearing its wake up words, just like everything else. If it were constantly streaming they would be caught as soon as someone decided to fire up wireshark (i.e. very soon after release.)
  - joshumax 9y ago
    I even ripped apart an Echo dot a little while ago, de-soldered the NAND from the device and dumped the FireOS-based ROM to take a peek myself. It is indeed performing audio processing for the wake-up word entirely locally.
    - imglorp 9y ago
      This is the most important comment here. Some independent verification. Worthy of a blog post?

monkeynotes 9y ago

Well, it looks to me that all we have at the moment to get upset about is some speculation. We need to make a bunch of assumptions to get to 'Google is logging everything we say'.