If they batch compress and upload, it wont be a stream. It could be aes-256 lzma2 compressed chunks. Using the right codec, they could likely upload a hour of audio in a few hundred kbytes. I have seen some demos of highly bw optimized codecs.

Here are some staring points. [1] I would go with Opus. My voice chat server uses that.

[1] - https://stackoverflow.com/questions/167533/best-voice-compre...

> Using the right codec, they could likely upload a hour of audio in a few hundred kbytes.

But at the same time, you're convinced that you would notice if your phone was slipping out these kilobytes as part of its usual communication?

It seems to me that you've just arbitrarily picked a level of paranoia where you've decided that this can definitely happen with the Alexa, but won't happen with your phone. Your proud paranoia isn't applied consistently.

I am not convinced of anything.

The risk of the throw away device uploading my audio is significantly lower than a device powered in my home, with dedicated fast internet.

What if they convert speech to text like the American intel agencies do as proved by the Snowden and Vault 7 leaks?

That's possible but that's quite a bit of processing for actionable speech to text which means you have a chance of the device acting sluggish when you query it or you have to give the device way more processing power than it needs which increases cost and from my understanding with devices like Amazon's, they're already being sold relatively near cost because they are intended to get you to spend more money inside the Amazon ecosystem.