[jmholla]

> Someone would have noticed if all the phones on their network started streaming audio whenever a conversation happened.

You don't have to stream the audio. You can transcribe it locally. And it doesn't have to be 100% accurate. As for user identify, people have mentioned it on their phones which almost always have a one-to-one relationship between user and phone, and their smart devices, which are designed to do this sort of distinguishing.

[Mogzol]

Transcribing locally isn't free though, it should result in a noticeable increase in battery usage. Inspecting the processes running on the phone would show something using considerable CPU. After transcribing the data would still need to be sent somewhere, which could be seen by inspecting network traffic.

If this really is something that is happening, I am just very surprised that there is no hard evidence of it.

[energy123]

They wouldn't do full transcription, it'd be keyword spotting of useful nouns ("baby", "pain", "desk", etc).

The iPhone already does this when you wake it up with Siri.

[caminante]

Even the parent's envelope math is approachable.

With their assumptions, you can log the entire globe for $1.6 billion/day (= $0.02/hr * 16 awake hours * 5 billion unique smartphone users). This is the upper end.

[htrp]

Terrifying cheap if you think about it