You mean... A tiny camera? A photoresistor is a device which measures light hitting it. It's a 1-pixel camera. In exactly the same way a decibelmeter is a little microphone.
My point is, OP is looking for some complicated algorithm to detect and eliminate speech, but you don't need that - you can just eliminate pretty much everything in the first place.
> Why not? It's a trivial[1] amount of data after you apply a simple noise gate and a low bandwidth codec.
5.6kbit/s to record noise levels is like using a CCD camera and a microprocessor for image detection. You could sample the noise level once every 10 minutes, store it in a uint32, and you'd get 5.6kbit/hundred days. You could ship the data over an SMS link. The units could be battery powered and last months.
Why would you use as camera as the sensor for simple presence detection? An IR LED+photoresistor is much cheaper.
> No one ever worries about the privacy implications of this
Yes, I certainly would be suspicious of both the camera and the attached microprocessor needed to "decide if it's light or dark" from "image data". Anybody that chooses to pay for that kind of massive over-engineering instead of using a cheap IR emitter+detector is trying to do more than just flush the toilet.
> Is Google going to record thousands of hours of sidewalk microphone recordings and store them somewhere?
Why not? It's a trivial[1] amount of data after you apply a simple noise gate and a low bandwidth codec.
[1] https://news.ycombinator.com/item?id=16346416