[sirtastic]

Late to the game here and apologize if this has already been mentioned but may shed some light on what is actually stored. A while back facebook officially announced what they "store" in regards to what they record when the app is open and has access to your microphone. The app itself turns your voice audio into a profile that is then sent to the server, it does not send your actual audio. This light(er) weight audio profile is then matched to words which are then used for marketing. I wouldn't be surprised if Echo, Google Home, etc do something similar since sending the full audio recordings would be both bandwidth and storage heavy.

[kobeya]

"Audio profile" meaning MFCC coefficients? That's still enough to fully reconstruct a meaningful audio stream. Obviously, since it's enough to do speech recognition.

[fucking_tragedy]

What difference does it matter if you can still construct words and thus the content of your speech?