Your most solid guarantee that your device isn't constantly spying on you is a battery.
Listening for words is power-intensive.
Listening for words is power-intensive.
For me the most solid guarantee is the lack of a business case. Why risk the huge fallout of when this spying is inevitably uncovered when the profit to be had of listening in to people’s private conversations at scale is small. Individual private conversations might carry huge value, but discerning those from casual conversation at scale is next to impossible.
https://ryanplant.net/love.wav
Its only mistake was missing the "you" in "you want to be."
Codec2 remains fairly clear down to human ears down to 2.4 Kbit/sec but I had no luck getting it recognised, even with much simpler and clearer samples.
This individual anecdote tells you nothing about the state of the art, but I wanted to note it anyway because I was astonished at how well 6 Kbit/sec is handled.
But the love.wav gives a bit rate of 768kbps, 6 sec of audio at 571 KB storage size.
Another poster also did the math on 5.6 kbit/s (700 bytes/s).
> 5.6 * 60 * 60 / 1024 / 8 = 2.4609375 MB/hr
But should that 5.6 not be 768 to get love.wav voice quality? What am I missing like a moron?
Thanks for the example and anecdote, something to actually grasp in trying to understand the problem.