Hacker News Reader

Comment by cbr | Hacker News Reader

Reasons to think it's client side:

* Sending all microphone input to a remote server would be really bad for battery life and data usage.

* It still works even in airplane mode.

* On ChromeOS (which isn't Android, but may share the same code) their privacy whitepaper [1] says:

    If you opt-in to the feature, Chrome OS will listen for
    you to say "Ok Google" and then send the audio of the
    next thing you say, plus a few seconds before, to
    Google. Detection of the phrase "Ok Google" is performed
    locally on your computer, and the audio is only sent to
    Google after it detects "Ok Google"."

[1] https://www.google.com/chrome/browser/privacy/whitepaper.htm...

View on HN

Also, keyword detection algorithms are different from general purpose speech transcription algorithms. The general pattern is to write a specialized model that's really good at robustly identifying the key phrase, run that continuously, then trigger recording and full transcription only when the key phrase is heard.

Those keyword identification algorithms can be implemented at the hardware level. I believe this is done for specialized devices like Echo and Kinect, not sure how phones do it.

The 1st generation Moto X had a special minimal chip used just to detect you saying "Ok Google Now" (or a custom phrase) to turn the phone on when it was sleeping. But I think phones that only detect a start word when the screen is on probably don't have custom low-power hardware for it.

Sending all microphone input to a remote server would be really bad for battery life and data usage.

It still works even in airplane mode.

I agree that they probably only send speech recorded after the activation words. However, I don't arguments think your arguments hold up. Speech data compresses exceptionally well and generally requires lower sampling rates and/or resolutions. So, it's feasible to compress the speech signal (possibly hardware-assisted), buffer it, and send it to the server at some regular interval or when the user wakes the phone.

Yes, it is possible for them to be sending all recorded sounds (even if the hotwords are not detected) to a remote server. However, as discussed above, there would be no legitimate reason for doing so—it is demonstrably true that the phone is able to recognize the hotwords even when a remote server is not reachable.

If you're only interested in spying, sure. If you want the phone to respond when a user says "ok, google" rather than a minute later it wouldn't work.

Reasonable, thanks. Question is if I said "ok ..." in a noisy environment or my accent is awful or I'm eating an apple, will android try to reqognize the phrase on server or not? If so, phrases similar to "ok, google" may be saved by google.

Like just the other day, I was having a conversation with my friend Snu. I said, "Okay Snoogle, it's time to dispose of the bodies of that couple we killed downtown last night.

It may specify it within their patent [1]; "Contextual advertising for video and audio media " .

[1] https://www.google.com/patents/US9087331

> * Sending all microphone input to a remote server would be really bad for battery life and data usage.

Not if you transcribe it first, and upload compressed text.

Reply on news.ycombinator.com