But I've always thought this was not about convenience for the user but for easier access to datamining yes
Like I can't imagine it has a huge bandwidth (outside of bt range) or amazon would monetize it as a actual network, but I would still like to know (order-of-magnitude) how many bits of data per unit time it can handle and how much information can we pack into that bandwidth.
Any spying is terrible, but there is a big difference between "ids of what shows you watched yesterday" and "compressed audio clips from 20 minutes ago".
EDIT some information: 500MB per account per month & 80kbps per bridge device ... https://www.aboutamazon.com/news/devices/everything-you-need...
If they do voice recognition first then send the recognized text, that cuts down on the required bandwdith significantly.
That said, (puts on black hat) If you are trying to tell when someone is open to impulse buying out sadness or anger or whatever, voice information is probably better than just doing sentiment analysis of the text.
On a low bandwidth network, getting at that same information means you'd have to try to running the analysis locally on the hub device. A device that is, in theory, servicing many clients. Maybe the device can manage TTS, but it seems less likely to be able to run a large ML model. Plus bandwidth limitations means amazon is less likely to get that information until later than they would from an Alexa device connected directly to the internet.
My goal was to gauge how bad of a security nightmare sidewalk is, not be an apologist for it.