Alexa Guard was rolled out in the US. This is a huge technical accomplishment and a lot of Amazoners are likely high-fiving each other other. It’s not available outside the US yet.
There were a few ways to implement this. One way was to constantly stream audio data to Amazon’s servers where the data was processed. This is probably not how Amazon is doing this. It would be a huge amount of data and processing (tens of millions of Echos all doing sound recognition). Instead, it’s likely that Amazon has created a local / on device service that listens for the desired sound (glass break is high energy). If it works similar to the Alexa trigger, than it likely streams that buffered audio to its servers where it does a secondary verification against a much larger dataset.
The benefit of this hybrid approach is that you can have a much higher rate of false acceptance and the service will still work well. You’d rather not miss glass breaking if you’re saying that’s what you’re listening for. In case of a false detection, the server will just shut down the stream.
More of these devices will be cheaper and more capable. With everything monitored, all devices with trackers, does this mean that theft becomes more costly / riskier to the thief? The end of petty crime might be near.