Hacker News new | ask | show | jobs
by camdat 1134 days ago
It would be trivial to confirm yourself, Wireshark can scan for outgoing network packets and audio would require a large, continuous, byte stream.

Unless you assume that Facebook et al are running ASR models on-device all the time and somehow making it invisible to the end-user.
3 comments
antognini 1134 days ago
The main problem with the ad-companies-are-listening-to-you theory is that audio processing is very power hungry. Running an ASR model locally would eat up a ton of power. Just doing wakeword detection (where you're only listening for a specific phrase like "Hey Siri") generally requires a dedicated specialized chip so as to not impact power consumption too much.

Same problem if they were surreptitiously streaming audio to their servers. You would see it from the outgoing packets and streaming that amount of data would also be fairly expensive.

link
pcdoodle 1134 days ago
What if it's not audio processing real time?

Some voice codecs get their rate down to 2-3kbps. Maybe it could store and forward with content loads without raising flags?

Just spit balling here.

link
m3047 1134 days ago
Audio doesn't require all that much bandwidth. 20Kb/Sec should be plenty for reasonable fi. "CD quality" would be 40Kb. Ultrasound maybe double. So... 20Kb would be 20 packets per second and that's being generous. That's without compression.
link
sgerenser 1134 days ago
20 packets per second continuously from your phone to a single server, when you’re otherwise sitting there doing nothing would be very noticeable (for anyone who cared to look for it). And yet I’ve never seen any evidence reported of this.
link
anon223345 1134 days ago
Good call, I’m going to try this actually
link
camdat 1134 days ago
I can save you the effort, it's not happening. This would be a monumental national security concern for non-US states, not even considering the technical limitations that exist for storing or utilizing literally millions of audio samples a minute.

People are incredibly predictable using only a handful of demographics, there's simply no need to invest the astronomical amount that would be necessary to process these conversations when there's already many simple ways to track/generate user interest.

link